DETERMINANTS OF RURAL HOUSEHOLDS DEMAND FOR MICRO HEALTH INSURANCE PLANS IN TANZANIA. ARNOLD KIHAULE Ardhi University Dar es Salaam, Tanzania

Transcription

1 A 6 DETERMINANTS OF RURAL HOUSEHOLDS DEMAND FOR MICRO HEALTH INSURANCE PLANS IN TANZANIA ARNOLD KIHAULE Ardhi University Dar es Salaam, Tanzania WORK IN PROGRESS REPORT OCTOBER, 2013

2 REPORT SUBMITTED TO AFRICAN ECONOMIC RESEARCH CONSORTIUM Abstract The study analyses factors determining rural households decision to purchase micro health insurance plans in Tanzania. In the country, only a few rural households have health insurance cover, although insurance schemes that caters for those who have low income or poor are available. Thus, many rural households make out of pocket payments for health services in the episodes of illness or injury. Given that 34 percent of rural households are poor, out of pocket spending aggravates poverty and restricts some of them from using health services, when ill or injured. Several scholars have analyzed socio-economic determinants of rural households decisions to purchase health insurance plans in Tanzania. Little attention has been given to the outcomes of purchasing micro health insurance plans in regard to the utilization of health services and protection against catastrophic health spending as the determinants of demand for micro health insurance plans. This is the gap this study seeks to fill. The study adopted the probit model and matching estimator methods to analyze the outcome of rural households membership in the micro health insurance plans on the utilization of the health services in the episodes of illness or injury. The data from Tanzania Demographic and Health Survey of 2011 were used to analyze households behavior. A comparison of the utilization of health services among the members and non-members as well as poor and non poor households in the episodes of illness or injury was also done. The estimation results revealed that the membership in micro health insurance plans resulted in an increased utilization of the health services among the poor households and not among the non-poor. In addition, health insurance plans did not provide households with protection against catastrophic health spending, when ill or injured. The probable reasons are that members do not save when visiting health facilities and that the plans provide limited range of benefits to households. Thus, it is recommended that the health facilities should provide adequate medical supplies in order to minimize out of pocket spending for rural households, when ill or injured. Also the government need to subsidize micro health insurance schemes so that they provide a wide range of services to members. 1

3 1. Introduction 1.1 Background to the study This study examines the factors determining rural households decision to purchase micro health insurance plans in Tanzania. The enrolment of the rural households in the micro health insurance schemes enables them to easily access modern health care services in the episodes of illness or injury. The accessibility of rural households to health care service in episodes of illness or injury is an issue of concern to policy makers. The reason is that large expenditure for health care services in episodes of illness or injury leads to expenditure shock to households and it aggravates poverty (Spaan et. al. 2012, Wilms 2006). means In this regard, a number of rural communities have initiated micro health insurance schemes as the means to facilitate access to health care services in the episodes of illness. It is also a means to help to spread health risk among the members and protect households against expenditure shocks in the episodes of illness or injury. Micro health insurance schemes are organised either along economic interest groups or religious denominations (Jutting, 2003, 2001). The established micro health insurance schemes also facilitate the mobilization of revenue for financing health services. The issues of inadequate resources in the health sector and households accessibility to health care services have been a concern of governments and international organisations as well. The Bamako Initiatives launched in 1987 by WHO and UNICEF together with the governments of a number of countries, adopted a new orientation in the delivery of health care services and health policy in order to improve households access to health care services (Kouadio, Monsan, Gbounge 2008, Bitran and Yip 1998). Furthermore, WHO in 2006 passed resolution on universal coverage of health insurance among the population in different nations in order to promote equity in accessing health care services. The Resolutions issued aims to promote individual households access to health care services. In this regard, the Tanzanian government had introduced a number of policies and programmes with the aim of not only improving households access to health care services, but also ensuring sustainability in financing the health sector. Specifically, in Tanzania economic policy reforms undertaken in 1986 have led to the commercialization of the health sector and inclusion of the private sector in the provision of health care services to the public among other things (Kida 2012). 2

4 In addition, in 1994 the government initiated health sector reforms that aimed at improving equality in utilizing and accessing health care services among the Tanzanians. The reforms also focused on achieving efficiency in health care service delivery system. As a result, the cost sharing system in health care services delivery was introduced. In order to improve the delivery of health care service, the government also introduced the National Health Policy in 2003 which was revised in Among the objectives of National Health Policy is to ensure that health care services are available to all people in both urban and rural areas. In addition, Tanzania Health Sector Strategic Plan III of and the preceding one have emphasized equity in accessing health care service among the populace (URT 2009). The policy and programmes have aimed at enrolling at least 30 percent of Tanzanians in the health insurance schemes. The introduction of health insurance in Tanzania dates back to 1997, when the National Social Security Fund (NSSF) introduced Social Health Insurance Benefit (SHIB) schemes for the formal and informal sector employees. In 1999, Community Health Fund (CHF) was introduced mainly for those in rural areas and in the informal sector (Minja et.al. 2008, URT 2001a; 2001b). National Health Insurance Fund (NHIF) that caters for civil servants was introduced in 2003 (National Health Insurance Fund, 2003). In addition, as pointed out earlier, in some areas, local communities have established and operate private self managed micro health insurance schemes to cater for informal sector employees or rural households. Notwithstanding efforts to introduce health insurance schemes in the country, only 15% of populace have health insurance cover. Most of those with health insurance plans are in urban areas. Consequently, a large number of rural households make out of pocket payments when visiting health facilities in the episodes of illness or injury. The out of pocket payment for health care services limits households access to modern health care services when sick. A number of authors provide evidences on the exclusion of the rural households in accessing health services in episode of illness or injury as a result of inability to pay for the services (Kida 2012; Mushi 2007, 2004b; Xu, Evans, and Kawabata 2003; Jütting, 2001). However, there is a number of households who make out of pocket payments for health care services in the episodes of illness or injury. The out of pocket spending for health care is estimated to be 80 percent of the total private spending and 50 percent of total health care expenditure in Tanzania (NBS and Macro International Inc., 2007). This is a large proportion of the incomes especially for the rural households given that 34 percent of them are poor. Thus, health insurance schemes can facilitate rural households 3

5 access to health care services and provide financial protections against expenditure shocks in the episodes of illness or injury. Membership in the health insurance schemes is also expected to contribute in overcoming poverty. Such outcomes are expected to motivate rural households to enrol in micro health insurance schemes. Nonetheless, a number of authors had questioned the envisaged outcomes of rural households membership in the micro health insurance scheme (Shimeles, 2010; Mushi, 2004a; Jütting, 2001). The reason is that most micro health insurance schemes have fixed premiums or limited range of premiums that the poor households cannot afford. In addition, the members of the schemes, incur extra expenses arising from travelling to modern health care facilities and purchasing prescribed drugs in the episodes of illness or injury. Notwithstanding the arguments provided by different authors, micro health insurance schemes are useful in facilitating access and equity to using the health services among the low income and poor households. It is argued that provided proper arrangement are available even the low income households can afford to purchase health insurance plans. The equity in utilizing modern health care services is linked to households or individuals ability to get access to health care services in the episodes of illness or injury. The Household Budget Survey of 2007 revealed that 69 percent of households or individuals who felt sick visited modern health care facilities (URT 2009). The survey also revealed that one third of household members who felt sick or were injured did not visit health facilities. Inability to pay health care services and health facilities being far from their residence were among the reasons that restricted households access to health care services. Notwithstanding various arguments on the benefits of micro health insurance schemes, in developing countries including Tanzania, the focus of a number of studies on micro health insurance schemes have been on analysing socio economic determinants of households membership in the schemes. Little attention has been given on analysing the effects of membership on micro health insurance on the utilization of the health facilities and financial protection in episodes of illness or injury, as factors motivating household to enrol in insurance plans. In addition, only a limited number of studies have analysed the utilization of the health care facilities among the poor and non-poor insured households in the episodes of illness or injury in Tanzania. The above mentioned issues are the knowledge gaps this study seeks to fill. 4

6 1.2 The research problem In Tanzania, households out of pocket spending constitutes 80 percent of the total private expenditure and health outlays. Such a situation is undesirable especially for the rural households, given that 34 per cent of rural households are poor (NBS 2009). The large proportion of income used by households as out of pocket spending for health care service results in the reduction of the households disposable income and the capabilities to augment human and non-human capital (WHO, 2006; NBS, 2009). It also leads to the exclusion or limited utilization of health care services by households in the episodes of illness or injury. Nevertheless, in rural areas, micro health insurance schemes can bring about better health outcomes to the rural households by improving their access to quality health services, the protection against health risk and expenditure shocks in episodes of illness and injury. They can also enhance the financing of health care system and the delivery of quality health services to rural household as well. Evidences from some developing countries suggest that even the poor rural households are willing and able to purchase micro health insurance plans (Cook and Dzator, 2007; Kiwara, 2007, Ahuja and Jütting 2001, Jütting, 2001). Nonetheless, in Tanzania only a few households have purchased micro health insurance plans, leading to low pooling of revenue, the minimal spreading of health risks and limited protection against expenditure shocks in the episode of illness or injury. Such a situation aggravates poverty, in rural areas and limit the efforts to enhance the financing of health care system in the country. 1.3 Motivation for the study In the rural areas, voluntary micro health insurance schemes provide an opportunity to households to have access to health care system, to insure against health risk and reduce out of pocket spending in the episodes of illness or injury. They also provide opportunity to the households to improve their welfare. The reduction of poverty as envisaged by the Millennium Development Goals (MDGs) of 2000 and National Strategy for Growth and Reduction of Poverty (NSGRP) of 2010, depends on among other things households being able to protect themselves against expenditure shocks (URT, 2000, 2010). In this regard, rural households enrolment in the micro health insurance plans enables them to minimize out of pocket spending in the episodes of illness or injury and improve their welfare. However, despite the envisaged benefits of the micro health insurance schemes, few rural households have enrolled in them. 5

7 Previous studies on micro health insurance in Tanzania have analysed socio economic characteristics determining rural households enrolment in the health insurance schemes. Little attention has been given on the outcome variables, as among the factors motivating rural households to enrol in the micro health insurance plans. An insight into the effect of outcome variables on the membership in the micro insurance plans shall provide an additional knowledge on rural household demand for voluntary private health insurance as well as into their choices and actions in regard to health risk management. It shall also provide an understanding on strategies for promoting households enrolment in the micro health insurance schemes in rural areas in the country. 1.4 Main objective of the study The main objective of this study is to analyse factors determining rural households decision to purchase micro health insurance plans in Tanzania. Specific objectives The specific objectives of this study are as follows: (i) (ii) (iii) (iv) to analyse the effect of demand for health care services in the episode of illness or injury on households decision to purchase micro health insurance plans; to examine whether the need to protect households against catastrophic health spending in the episodes of illness or injury influences their decision to purchase health insurance plans; to analyse the utilization of the health care service facilities by households with micro health insurance plans and non-members in the episodes of illness or injury; To analyse the utilization of health care facilities among the poor and non-poor households who are members of micro health insurance plans. 2. Literature review on demand for health insurance and empirical studies This section provides the definitions of micro health insurance schemes and literature review on the demand for health insurance plans. It also presents empirical literature on the households decision to enrol in the micro health insurance schemes in different countries and Tanzania as well. 2.1 Definition of a micro health insurance fund In literature, a Micro Health Insurance Scheme (MHIS) or Micro Health Insurance Fund (MHIF) is also referred to as either a Community Based Health Insurance Funds (CBHIF) or Mutual Health Funds (MHF). The above mentioned schemes are considered as one type of private voluntary health insurance (Al-Khatib, 2007; Preker, 2004; Jütting, 2001; Jowett and Ensor, 2000). Inspite of the various terms used, the general consensus among scholars is that micro health insurance schemes or community based insurance funds are specific group organized insurance schemes, characterized by the pooling of revenue, 6

8 the sharing of health risk and voluntary membership. Thus, in this study, a micro health insurance scheme is also considered as synonymous to a community based health insurance and a mutual health fund. 2.2 Review of literature on demand for health insurance Demand for health insurance refers to the amount of insurance cover an individual is willing to purchase at different prices, that is, premiums (Feldstein, 1988). The appropriate amount of health insurance is purchased when the marginal benefit of health insurance is equal to the cost of insurance. The theories of large numbers and that of risk provide the foundations for the understanding of demand for micro health insurance plans. In particular, as per the law of large numbers, if a segment of population may fall sick and choose to pay premium for being sick, it will be possible to charge premium in advance to collect enough revenue to cover health expenditure with a profit margin to insurance firms (Pauly et. al., 2004). The underlying assumption is that individuals who are health risk averse can afford to pay health insurance premiums and to access health care services in episodes of illness or injury. In addition, the expected utility theory postulates that people purchase health insurance so as to hedge themselves against financial risks due to illness (Neumann and Morgenstern, 1944). Individuals purchase health insurance plans because they are risk averse and that they prefer certain losses over the uncertain ones of the same expected magnitude (Amponsah, 2009). In this regard, it is asserted that accessibility to health care and protection against health risk motivates individuals or households to enrol in the health insurance schemes. Furthermore, the theory of risk postulates that under the condition of rationality and risk averseness, the households decision to purchase insurance plans is made on the basis of expected utility to be gained (Bhat and Jain, 2006). The utility gained depends on the expected medical needs of the individuals or households in the episode of illness or injury and the level of financial protections. For the poor, households, accessing health facilities in the episodes of illness or injury and financial protection against expenditure shocks are expected to be the main factor motivating them to enrol in the micro health insurance schemes. In contrast to the theory of risk, the prospect theory suggests that individuals demand health insurance so as to transfer income from the healthy state to the ill state (Tversky and Kahneman, 1986). During the healthy state, medical care is an irrelevant good, while during the episodes of illness; it becomes a substitute to other goods and services. The underlying assumptions for the expected utility and prospects 7

9 theories of demand for health insurance are that households are risk averse and compare the expected gains from the health insurance plans over the cost of premium before purchasing health insurance plans. For those events with high probability of occurring, it is not worth buying health insurance because marginal benefit is less than marginal cost. In a number of both developed and developing countries an increase in health care costs has been observed (Pariyo et.al, 2009; Xu, Evans, Kawabata 2003). Consequently, governments have reduced the generosity of health care system and that of the social health insurance. The outcome of the reduced generosity has been the decrease in publicly provided health care services and increased co-payments. Such a situation is associated with the rise in households out of pocket spending on health care services and limited accessibility to health services. Health risk exposes the households to both direct and indirect costs, leading to an unexpected financial outflow that poses financial risk (Xu, Evans, Kawabata 2003). Given that health risk may lead to a higher out of pocket spending households respond to such a situation by adopting different choices. They include either an increase in pecuniary saving or change in financial allocation or enrolling in the health insurance plans. As far as demand for health insurance is concerned, households have two choices. The first is to purchase health insurance plans and incur small loss in the form of premium and the second one is to decide to self insure, that is, facing a small possibility of large loss in event that illness or injury will occur, or large possibility that medical loss will not occur (Feldstein, 1988). Thus, the demand for private health insurance is the predictor of enrolment in the insurance plans. It is believed that rural households make the assessment of health risks and associated expenses before enrolling in the health insurance schemes (Lammers and Warmerdam 2010). In this respect, households or individuals enrol in the insurance schemes to overcome income and expenditure shocks in episodes of illness or injury among other things. Given that micro health insurance is one type of the voluntary private health insurance, the theory of demand for private health insurance has been used as the framework for analysing demand for micro health insurance in developing countries (Al-Khatib, 2007; Ahuja and Jütting, 2002; Jütting, 2001;2003). The theory of demand for health insurance provides a framework for analysing demand for micro health insurance schemes as well. The reason is that the theory considers health insurance as a risk avoidance mechanisms and a means to ensure individuals or households access quality health care services in the 8

10 episodes of illness or injury. Specifically, the theory considers the outcomes of membership in the health insurance scheme, the issues of interest in this research. 2.3 Empirical studies on households demand for micro health insurance In a number of studies on households demand for micro health insurance in developing countries, the focuses have been either on analysing the demand or supply side factors or both of them. In particular, a number of authors have examined factors determining low income or rural households enrolment in the micro health insurance schemes in developing countries (Ranson et.al., 2007; Basaza et.al., 2007; Jütting, 2003; 2001; Ahuja and Jütting, 2002; Osei-Akoto, 2001). In addition, the authors used different variables and methods to analyse demand for micro health insurance plans. The findings from various studies had revealed that households and community characteristics, that is, age, of the head of household, gender and income were among the factors determining households membership in the micro health insurance schemes in rural areas (Lammers and Wermerdam, 2010; 2007; Jütting, 2003;2001; Ahuja and Jütting, 2002; Osei-Akoto, 2001). Other characteristics were education level of the household members, household size and religious denomination belonging. The same factors are expected to influence household demand for micro health insurance plans in rural areas in Tanzania. The reciprocal relationship in helping each other, among the family members including the extended families, has also been found to influence households demand for micro insurance plans (Basaza et.al., 2007; Jutting 2003, 2001;). Johar (2007) also had similar finding in his study on demand for micro health insurance in Indonesia. The relationship determines social values of individuals or households which guide the calculation of costs and benefits and the decisions to be members in the micro health insurance schemes or not (Basaza et.al. 2007). Specifically, the larger the social capital or network, the lower, the enrolment in the micro health insurance schemes is expected. Risk attitude of the households is another factor that influence households enrolment in the micro health insurance schemes (Kamuzora and Gilson 2007, Pauly et.al., 2004, and Preker, 2004). It is argued that in developing countries, individuals and households have low attitude towards health risk and this led few of them to enrol in the micro health insurance schemes. Kiwara (2007) had a different view on the issue and argue that in various parts of developing countries, where appropriate mechanisms are available low income or rural households have joined micro health insurance schemes. This suggests that even the low income households can purchase micro health insurance plans provided that appropriate arrangements for them are available. 9

11 A number of scholars have examined the effect of the supply factors, that is, the quality of health care services, the availability of health care facilities and waiting time on the demand for micro health insurance plans in developing countries (Basaza et.al. 2007, Johar 2007; Ranson et. al Bhat and Jain, 2006;). Other supply factors that scholars had examined are the benefit packages, the insurance plans provide, the degree of freedom to choose providers and the extent of compensation provided by the insurance schemes (Zweifel and Breyer, 1997; Sanhueza and Ruiz-Tagle, 2002). It is expected that the above mentioned supply factors also determine households decisions to purchase micro health insurance plans. In particular, factors such as the availability of health care services and the benefit packages have a bearing on the household spending on health services in the episodes of illness or injury. In this regard, they also influence the utilization of health care services and the decision to purchase health insurance plans. Shimeles (2010) and Spaan et. al. (2012) examined factors influencing households enrolment in micro health insurance schemes. Among the factors considered were the outcomes of membership in the micro health insurance schemes, that is, households utilization of health care services in the episodes of illness or injury and protection against catastrophic health spending. Shemelis (2010) also examined the use of health care facilities among the poor and non poor households, who were members and non members of the insurance schemes in order to ascertain whether they had different pattern of the utilization of health services when sick. Overall, the studies of Spaan et. al. (2012) and Shimeles (2010) have focused on the issues of equity, access and inclusion of the poor in utilizing health care services in episodes of illness or injury. The issues are also worth examining in the context of Tanzania, since a small proportion of rural population has purchased health insurance plans. In sum, a review of literature on empirical studies on demand for micro health insurance plans in developing countries reveals that researchers have focused on either examining socio economic or supply factors or both as determinants of household membership in the schemes (Kiwara, 2007; Basaza et.al., 2007, Bhat and Jain, 2006 Ahuja and Jütting, 2002; Jütting, 2003). Few studies had analysed the effect of membership in the micro health insurance schemes on the utilization of health services among the members. Furthermore, little is known on the outcomes of the membership in micro health insurance schemes in protecting the households against catastrophic health spending in episodes of illness or injury (Shimeles, 2010; Spaan et. al. 2012). In addition, few studies have examined whether rural households needs to utilize health care services and get protection against catastrophic health spending in episodes of illness or injury motivate them to purchase micro health insurance plans. 10

12 2.4 Demand for micro health insurance plans some methodological issues Experimental methods, cross sectional studies, case studies and econometric methods are among the approaches that have been used to examine the determinants of demand for micro health insurance plans in developing countries (Shimeles 2010, Kiwara, 2007; Ranson et. al., 2007; Bhat and Jain, 2006; Ahuja and Jütting, 2002; Osei Akoto, 2001; Jütting, 2001). The methods have a number of advantages and shortcomings. In this section only cross section survey and econometrics method are discussed. Cross Sectional Survey The cross section survey methods have been used to analyse the demand for micro health insurance plans in a number of studies in developing countries (Bhat and Jain, 2006; Osei Akoto, 2001; Ahuja and Jütting, 2002, Kumar 1999). The approach is useful if an overview of a particular phenomenon at one time and across the population is sought. One of the shortcomings of the cross sectional survey method is the selection bias in the process of collecting data to be used in analysing demand for micro health insurance. The problem may be addressed by using instrumental variables as suggested by Heckman (1979) and matching estimator as recommended by Shimeles (2010), Johar (2007), Todd (2006), as well as Smith and Blundel (1986). In addition, in cross sectional studies, econometric methods, such as probit or binary logit models have been used to analyse the determinants of households membership in micro health insurance plans (Shimeles, 2010; Bhat and Jain, 2007; Ahuja and Jütting, 2002; Osei Akoto, 2001; Ahuja and Jütting, 2002; Jütting, 2003). One of the problems encountered in estimating the demand for micro health insurance plans using probity and binary logic models is inability of the models to address the problem of endogenity of the variables used and the selection bias. The problem arises because the data are not obtained from the randomised experiments and selection bias is inherent among the members joining micro health insurance schemes. Such a problem leads to implausible results. In ascertaining the effects of membership on micro health insurance schemes on demand for health care services and income protection, regression methods had also been used (Bhat and Jain, 2007; Ahuja and Jütting, 2002; Osei Akoto, 2001). However, as noted by Heckman (1978) in such models both dependent and independent variables are discrete. Thus, the estimation process tends to be complicated. In order to overcome the endogenity of the variables used and selection bias, Shemelis (2010) suggested the use of good instruments that impact on health and income only through the membership in the micro health insurance schemes. This requires undertaking the tests for weak exogenity (Smith and Blundel 1986). In case the residuals from the first stage regression including the instruments exhibit weak 11

13 exogenity, then one can use probit model to obtain coefficients that impact on outcome variables. Similar, approach shall be used in this study to overcome the endogenity of variables and selection bias. In order to ascertain the impact of micro health insurance on the utilization of health care services among the members and non members as well as poor and non poor households, matching estimator methods had been used in a number of studies (Shimeles 2010 and Jahor 2007). The method also enables the estimation of the outcomes of membership in the health insurance plans. The outcomes referred to include households utilization of health care services and protection against catastrophic health spending in the episodes of illness or injury. Similarly, matching estimator methods have been used to compare the utilization of health care services among the members of micro health insurance and non members. The matching estimator has been used together with the probit models. 2.5 Empirical studies on demand for micro insurance plans in Tanzania A number of scholars have analysed various issues concerning the government supported community based health insurance schemes and those which are privately managed in Tanzania (Kiwara, 2007; Kamuzora and Gilson, 2007 Mushi, 2004a; 2004b) In particular, scholars have examined different issues in regard to the impact of the micro health insurance schemes on the accessibility to health care services and on households consumption, affordability of premiums and saving in the episodes of illness or injury (Kiwara 2007; Mushi 2007, 2004a, 2004b). Kiwara (2007) examined whether group premiums or individual premiums were better alternatives for making micro health insurance plans affordable to households in the informal sector. The author found that the modes of paying premiums had an impact on households decisions to enrol in the micro health insurance schemes and the continuity of the households membership in the schemes. A similar finding was observed by Mushi (2004b). Thus, premium or price of insurance plan determines a household purchase of the plans. Furthermore, Lamers and Warmerdam (2010) argue that, individuals or households purchase health insurance if the expected utility when insured is larger than when not insured. This suggests that households purchase health insurance plans because they are guided by preference for certainty and wealth security to uncertainty. Mushi (2007) also observed that rural households decisions to join community health funds was determined by whether they were going to save in the episode of illness or not. In case households save by being members in the insurance schemes in the episode of illness or injury, they were motivated to join in the health plans. This also suggests that financial protection against catastrophic health spending 12

14 influenced households memberships in the micro health insurance schemes as pointed out in literature (Lammer and Warner 2010, Dror et. al, 2007). Furthermore, in studies on Tanzania, different research methodologies have been used to examine government supported community health funds or micro health insurance schemes (Mushi, 2004a; 2007; Kiwara 2007; Kiwara and Heijis, 2001). Specifically, Kiwara (2007) used experimental approach to examine the effect of the introduction of the different modalities of paying premiums for micro health insurance schemes on the membership enrolment in Dare es Salaam. In addition, Kiwara and Heijis (2001) used exploratory cross section method to examine the feasibility of health insurance in the informal sector in Mbeya and Arusha in Tanzania. In analyzing the determinants of membership enrolment in the government supported community health funds, Kamuzora and Gilson (2007) used case study approach. Overall, a number of studies have examined the implementation of government supported community health funds in Tanzania (Kamuzora and Gilson, 2007; Kamuzora and Mushi 2004a,2004b). Some others had focused on analysing the feasibility of the micro health insurance schemes or their designs (Kiwara, 2007; Kiwara and Heijis, 2001). Few studies have examined the determinants of rural households membership in the micro health insurance schemes. Little is known about the relationship between membership in the private health insurance plans and the outcome variables in Tanzania. This study seeks to make the contribution on the issue. Furthermore, in contrast to previous studies on micro health insurance schemes in Tanzania, this study uses data from Tanzania Demographic and Health Survey conducted in 2010 and published in 2011 (NBS 2011). A simple probit model and matching estimator methods are adopted to analyse data. Probit model is used to analyse determinants of households membership in the private micro health insurance schemes. The matching estimator methods are used to examine whether membership in the micro health insurance scheme has impact on the households utilization of modern health care services as well as protection of members against health related consumption shocks and motivate them to enrol in the insurance schemes. The use of the matching estimator methods is prompted by the need to test the methods, using data from Tanzania in order to obtain plausible results. Similar methods have been used in other studies in developing countries (Shimeles, 2010; Jahor, 2007). 3. Framework of analysis and methodology The theory of demand for health insurance provides the conceptual framework for analysing households membership in the health insurance plan. In particular, it is assumed that individuals or households seeking health care services behave in economic and rational manner. The individuals or households 13

15 assess their health risks and the associated health expenses before they make decisions to purchase micro health insurance plans. This also suggests that health and financial risks motivate individuals or households to make decision to purchase health insurance plans or not to. As such households have two alternatives either to be a member of the micro health insurance scheme and incur a small loss or to decide to self-insure. In that respect, membership in the health insurance plans, facilitate individuals or households to access health care facilities in the episodes of illness or injury and get protection against catastrophic health spending when sick. In particular, the focus of the study is to analyse whether the outcome variables also influence households decision to purchase health insurance plans in rural areas in Tanzania. In order to realise the objectives of this study, econometric model that explains the relationship between households membership in micro health insurance schemes and the expected outcomes, that is, utilization of health care services and mitigation against health spending shock in the episodes of illness or injury is adopted. The model takes into account that membership in the micro health insurance plans is endogenous (Shimeles 2010, Smith and Blundell 1986). The model is specified as: y 1i = (y 2i γ 1 + x 1i β 1 + µ 1i > 0)..(1) y 2i = (x 1i + ٧ 2i > 0).(2) (µ 1i /٧ 2i) ~ NI ((0, (δ 2 1 / δ * 21 δ 12 / Σ 22 )) (3) Where x=x 1i, x 21 is a vector of observation on K=(K1+K2) and y 1 i and y 2 i are vector of dependent variables and endogenous regressor. Given that equations (1) and (2) are system equations, the estimation of the parameters can be done by using a bivariate probit model, if the assumption of normality of the error terms holds (Shimeles 2010). The two equations can also be estimated using two steps procedures and recursive full maximum likelihood as suggested by Heckman (1979). However, before estimating the equations, the test on weak exogenity between regressors and error terms need to be done. Equation (1) is to be estimated as bivariate probit model, once it is established that the instrument variables that has been determined, correlate with the regressors but not with the error terms. In addition, the estimation of the probit model equation shall enable the realisation of the first objectives of this study. In order to address the second, third and fourth research objectives highlighted in Section 1.3, matching estimator methods are used, as propagated by Rosenbaum and Robin (1983). The use of the matching 14

16 estimator methods entails organising data in the treated and control dichotomy conditional on observable covariates (Shimeles, 2010; Johar 2007; Todd 2006). The dichotomy allows the estimation of the three statistics for the purpose of evaluating the impact of membership in the micro health insurance plans on the utilization of health care services and protection against catastrophic expenditure shocks in the episodes of illness and injury. Specifically, the matching estimator methods entail estimating a number of statistics that help to identify the outcome of membership in the micro health insurance schemes. One of them is the Average Treatment Effect (ATE). The Average Treatment Effect (ATE) allows comparison of outcomes between the treated and the control sub group. The comparison is done by taking the randomly selected individuals from both sub groups so that the impact of micro health insurance schemes is evaluated. Another statistics estimated is the Average Treatment Effects of the Treated (ATT). The statistics is used to evaluate the programme impact among the randomly selected groups exposed to treatment and lastly Average Treatment Effect on the control group (ATC) is estimated. The statistic measures the impact of the micro health insurance schemes on randomly selected individuals or households within the control group, that is, those who are not members of micro health insurance schemes. The use of matching estimator method to evaluate the outcome of the membership in the micro health insurance requires the information on whether individuals or households are treated or not. The differentiation between the treated and control groups is done by using dummy D variable on the realised outcome, due to the treatment and Xi is used to represent the set of exogenous covariates used as control variables. In that respect, the definition in equation 4 is expected to hold. C C Y i = Y i (D i ) = Y i ; if D i =0; Y i if D i =1) (4) Xi = K x 1; coefficients of covariates for the members and non members of the insurance schemes The matching estimator method is used when a number of assumptions hold (Shemelis 2010 Smith and Blundel 1986). The assumption are as follows: (i) full information on the subject under study is available, and thus there are no unobserved factors, simultaneously correlated with outcome and the decisions to participate in the treatment, (ii) there is a positive probability of participation in the programme at all values of the covariate Xi The testing of the validity of the assumption shall be done before proceeding with the estimation of the models. 15

17 3.1 The estimation of the probit model As pointed out in Section 3, probit model is used to estimate the determinants of membership in micro health insurance plans in rural areas. The membership in the insurance schemes is specified as being determined by a number of variables as shown in equation (5). The model is specified as follows: P*=β 0 +β 1 DUi+β 2 HSPENDi+β 3 CONS+ β 4 X 1 + β 5 Χ 2 +β 6 CONS+ ε...(5) where P* > 0 if a household is a member of the insurance plan P*< 0 if a household is not a member The definition of variables is as follows: DU i = is a component vectors which include DU 1 = households utilization of health care services when sick DU 2 =households utilization of health care services for insured members DU 3 = households utilization of health care services of the poor DU 4 = households utilization of health care services for insured non-members DU 5 = households utilization of health care services of non-poor HSPEND 1 =households out of pocket spending for health care services (ratio of health expenditure to per capita income for all household) HSPEND 2 =households out of pocket spending for the poor (ratio of health expenditure to per capita income for poor) CONS=consumption quintile X 1 = vector of households characteristics (age of head of household, sex, education level, household size, occupation) X 2 = vector of community characteristics (geographic dummies). µ ~ N (0, δ 2 ) β 0, β 1 β 6 are parameters to be estimated Equation (5) shall be estimated after testing for endogenity and selection bias for the dependent variables. This shall be done by ascertaining the weak exogenity between the dependent variable and error term. In case weak exogenity assumption is rejected alternative estimation methods shall be adopted. 3.2 Matching estimator As previously pointed out in section 3, in order to realize the second, third and fourth objectives of the 16

18 study, estimations involving matching estimator methods shall be done. However, before conducting estimations, the assumptions guiding the use of matching estimator methods shall be tested if they hold. In addition, the operationalisation of propensity score matching estimators shall be done (Imbends, 2007). The aim of the operations is to remove biasness and obtain plausible results from the estimations done. 3.3 Testing of the assumptions: (i) Unconfoundness: This is a test whether there are no unobserved factors simultaneously correlated with outcome, that is, decisions to participate in the treatment. Y (0), Y(1) W (X)...(6) i Y (0) = α + β x i i + e...(7) i Y (1) = Y (0) + J...(8) i i Y i i = α + J. Wi + β xi + e...(9) i ei W Ix...(10) i i The equation (9) indicates that the membership or non membership (outcomes) is independent of the status of membership (W) conditional on characteristics of households (X i ). The tests for the assumptions shall be done using the following equation: Match: hhmemb DU 1, DU 2, DU 3, DU 4, DU 5, CONS, hspendhealth, households characteristics, (agehhod, hodeduca, household size, occupation), community characteristics,... (11) The variables are as defined in equation (5) (ii) Overlap 0 < Pr (W=1 X) < 1... (12) The expression implies that there is a positive probability of participating in the programme at all values of covariate Xi. The expression (12) implies that taking into account all the households, there are those who are members and those who are not members. A test shall also be conducted to ascertain whether the assumption holds before estimating matching estimators coefficients. 17

19 3.4 Propensity score estimator Before going further with estimations of the equations, it shall be necessary to test for propensity score. The aim is to ascertain the matching member of micro health insurance schemes and non-members using close similarity of characteristics of the households. The matching is done using proxies for utilization of health care facilities, as well as households and community characteristics. Given the equation (11), and high dimension covariates that characterise households, an attempt shall be made to remove biasness. This is done by adjusting the covariates into a scalar function (Imbends, 2007). The process is presented in equations (13) (15). Y (0), Y(1) W (X)...(13) then Y (0), Y(1) W e(x)...(14) E(x) = Pr (W=1 X=x)...(15) Following the estimation of the prospensity score, the sample shall be divided in sub-samples based on the value of the prospensity scores (Rosenbaum and Rubin 1983). This facilitates the estimation of the average treatement effects as the difference between average outcome for the treated and control groups. 3.5 Matching In determining matching for each treated (Y i ), and untreated groups (Y 0 ) respectively shall be determined, in an attempt to find out the effects of membership in the insurance schemes on the utilization of health services and protection against catastrophic health spending (Imbens, (2007). This is done for the treated, untreated and the controlled groups as specified in the equations (16) to (19) The equations (16) to (19) are presented below. Y ) Yi (1) = if wi = 1 and if wi = 0...(16) Yli Y ) Yi (0) = Yli if wi = 0 and if wi = 1...(17) Then the simple matching estimator shall be obtained using equation (19): 1 n ) ) sm = N = Yi(1) Y (0)...(18) i 1 After getting the units with nearest matches, 18

20 Estimation is to be done for µw(x) as well as for the modified matching estimators as follow: Y ) Yi = ) ) if Wi = 1 or if Wi=0...(19) Yli + u1( Xi) u1xj( i)) Y ) Yi = ) ) if Wi = 0 or if Wi=1...(20) Yli + u0( Xi) u0xj( i)) This will enable the determination of the unbiased matching estimator (Imbens, 2007). The new estimator is expressed as equation (21): bcm 1 n ) ) = N = Yi(1) Y (0)...(21) i 1 This estimation shall be done for both the treated, untreated observations as well as control group as detailed below:: (i) Overall Average Effect of the Treated (ATE) (ii) Average Treatment Effect for the Treated (ATT) (iii) Average Treatment Effect for the Control group (ATC) The estimations of equation (21) and the corresponding results are presented in Section Sample size and data In order to be able to estimate various equations provided in section 3 and realize the objectives of the study, it was necessary to determine the optimal sample size of data for the households, who are members and non members of the micro health insurance schemes. Equally, it was necessary to determine and define the variables used in this study. The details are provided in Sections 4.1 and Determination of sample size As pointed out before, the data used in this study were obtained from Tanzania Demographic and Health Survey conducted by National Bureau of Statistics in Tanzania, from December, 2009 to May The survey covered 475 clusters and 10,300 households NBS (2011). The study included men and women who were aged between and years old. While all women in the selected sample were interviewed, only one third of the men were included in the survey exercise. In addition, women had more questions to answer than men in the questionnaires administered, during the survey. 19

21 The objective of the survey was to collect data on various information for households in Tanzania. The survey collected data on households membership in micro insurance schemes and health service utilization in the episodes of illness or injury. Other data collected were those on households demographics and community characteristics among other things (NBS, 2011). The data of interest to this study were for the households in rural areas. Rural households were categorized into two sub-groups, that is, one group for members of micro health insurance schemes and non-members. In addition, the households were categorised into different income groups that is the poor and non-poor. The optimal sample size for members of the micro health insurance schemes and another one for non-members was determined by using statistical methods as shown in Appendix 1. Thus, given the sample frame of 6,266 rural households, the optimal sample size for the research was 362 households. This included 127 who were members of the micro health insurance schemes and 235 nonmembers who were chosen randomly from the sample frame. However, one of the limitations for using household budget survey data is the presence of indefinite factors that may limit the findings of the study (Shimeles 2010, Jütting, 2001, Heckman 1978). This is one of the limitation of the study. 4.2 Data and the definitions of variables As pointed in section 4.1, the Tanzania Demographic and Health Survey report of 2011 provided data for various variables that had been used in this study.the dependant variables of interests were membership in the micro health insurance schemes and non membership. It is deemed that membership in the micro health insurance schemes are influenced by the demand for health care services in the episodes of illness or injury. Thus, the dummy variable was used to capture the respective households who sought health care service in the episode of illness or injury and those who did not. Additional dummy variable was provided to capture for the utilization of the health care facilities for poor and non-poor and members and non-members as well. Similar approach had been used by Shimeles (2010). Another factor that is associated with the membership in the micro health insurance schemes is the protection of households against catastrophic health spending in the episodes of illness or injury for both poor and non-poor households and members and non- members as well. In this study, the dummy variables were used to capture households that experienced catastrophic health spending and for whom health spending was not burdensome. The variables are expected to be positively related to the membership in the schemes. 20