Health Insurance and Other Risk-Coping Strategies in Uganda: The Case of Microcare Insurance Ltd.

World Development Vol. 38, No. 3, pp. 369–378, 2010 Ó 2009 Elsevier Ltd. All rights reserved 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

doi:10.1016/j.worlddev.2009.09.004

Health Insurance and Other Risk-Coping Strategies in Uganda: The Case of Microcare Insurance Ltd. MARLEEN DEKKER African Studies Centre, Leiden, The Netherlands

and ANNEGIEN WILMS * Netherlands Authority for the Financial Markets, Amsterdam, The Netherlands Summary. — To reduce the burden of health expenditures in developing countries, health-insurance schemes have become popular and now feature prominently in poverty-reduction strategies. There is, however, limited empirical evidence on the effect of such schemes on the livelihoods of clients, especially regarding household strategies to finance medical expenditures. This paper explores the relationship between health insurance and other risk-coping strategies used to finance medical expenditures in Uganda. Insurance is associated with a lower frequency of asset sales but not with lower incidences of borrowing. The amount of money borrowed or generated through the sales of assets is lower for insured households. Ó 2009 Elsevier Ltd. All rights reserved. Key words — Uganda, Sub-Sahara Africa, health insurance, micro-insurance, poverty, risk-coping strategies

(Scheil-Adlung, Carrin, Ju¨tting, & Xu, 2006). This, in return, will increase the risk of ending up or being trapped in poverty. Bogale, Mariamand, and Ali (2005) and Krishna et al. (2006) demonstrate how the costs of illness contribute significantly to the impoverishment of households in rural Ethiopia and Uganda, respectively. In this context, reducing the financial burden of health care, for example, by having health insurance, allows earlier medical treatment and enhances a household’s long-term welfare as it may well shorten the duration of an illness, reduce the number of workdays lost, and improve productivity at work through having better health (Jutting, 2004; Young, Mukwana, & Kiyaga, 2006). More indirectly, when households are better protected against high medical costs, they are less likely to have to rely on other risk-coping strategies and may be able to accumulate savings and assets, thus improving their general welfare. In line with these arguments, policy debates have placed a great deal of emphasis on the development of health-insurance products for the poor. In the past two decades, hundreds of small schemes have been implemented across the globe (Bennett, Creese, & Monash, 1998) but, to date, profound

1. INTRODUCTION Achieving the Millennium Development Goals (MDGs) remains an important global challenge. Better protection for the poor against health risks is crucial in this endeavor and micro or community-based health-insurance schemes are being advanced as a means to reduce and stabilize the costs of treatment, increase access to health care and to reduce the use of costly risk-coping strategies (IFC, 2009; ILO, 2008; WHO, 2006). Previous studies have shown how insurance increases health-seeking behavior and reduces out-of-pocket (OOP) expenditures for medical treatment. Possible reductions in the use of other coping strategies, as an indirect effect of health insurance, have not been addressed in much detail. Based on data from specific areas in Uganda, this paper explores the relationship between health insurance and the use of other strategies to finance health care. Illness is a significant risk for people in developing countries (see, e.g., Dekker, 2004; Dercon, Hoddinott, & Woldehanna, 2005; Leliveld, 2006) and can have considerable short-term financial effects on the household affected. Illness is likely to reduce a household’s income if people are not able to work and may also result in additional expenditures to cover costs of treatment. As it is not uncommon for people to lack the cash to pay for medical fees (Asenso-Okyere, Anum, OseiAkoto, & Adukonu, 1998), people may forego treatment, with potentially detrimental effects for their long-term health. Alternatively, households use costly risk-coping strategies to pay for medical care: they reduce spending on basic needs, sell household or productive assets or borrow money. In a study of coping strategies in Uganda, Leliveld (2006) reported how households sold land, cattle, or goats or used their savings to respond to (long-term) illness. Such strategies are expensive and may endanger the future economic status of the household by depleting its finances through indebtedness and its future income-generating capacity by selling productive assets

* The field research for this paper was conducted by Annegien Wilms and Matthijs Verweij and invaluable research assistance was provided by Francis Somerwell, Christina Makobole, and Godfrey Asiimwe (Microcare, Uganda). The Department of Development Economics and the Dittmar Fund of the Vrije Universiteit in Amsterdam kindly provided financial assistance for the project. The authors would like to thank Andre Leliveld, Jan Willem Gunning, everyone who attended the conference on ‘‘The Pursuit of Certainty: Applications in Measuring Poverty” at the Maastricht Graduate School of Governance and the five anonymous referees for their comments on earlier versions of this paper. The views expressed here and any unintentional errors are the authors’ and do not represent any official agency. Final revision accepted: July 23, 2009. 369

370

WORLD DEVELOPMENT

empirical evidence of the effects of health insurance is still limited (Ekman, 2004; Jutting, 2005; Scheil-Adlung et al., 2006; Young et al. , 2006). Many studies describe the institutional underpinnings and the performance of the provider (see, e.g., Atim & Sock, 2000; McCord, 2000) or discuss the potential contribution of micro insurance for financing health systems (Dror et al., 2005). 1 Some have documented the impact of insurance schemes on health expenditure and treatmentseeking behavior and have found that health insurance increases health-care utilization (Ekman, 2004; Preker, Carrin, Dror, & Jakab, 2001; Schneider & Hanson, 2006). Jutting (2004), Jowett, Contoyannis, and Vinh (2003) and Ranson (2002) have all reported reduced levels of OOP payments for medical treatment as a result of insurance. Information on the relationship between insurance and the use of other coping strategies is scantier. Apart from a descriptive comparison of the incidence of risk-coping strategies in the context of three different health-insurance schemes (Scheil-Adlung et al., 2006), the relationship between health insurance and the sale of assets or borrowing money to pay for medical expenses has not been tested statistically and remains unsubstantiated. 2 Based on household-level data collected in five rural and two urban communities in Uganda where the private health-insurance provider Microcare Insurance Ltd. (Microcare) operates, this paper considers the relationship between insurance and OOP expenditures, and then addresses the correlation of insurance and the incidences and value of asset sales and loans. We have found that health insurance is associated with lower OOP expenditure on health and with less use of other riskcoping strategies. This association between health insurance and other risk-coping strategies calls for additional empirical work on the relationship between insurance, risk-coping, and poverty. If reductions in other coping strategies can indeed be attributed to health-insurance schemes and this is found in a wider range of schemes, health insurance can be seen as relevant beyond its direct effects on health-seeking behavior and reduced health expenditures, and has important indirect effects on household well-being too. This paper is organized as follows. Section 2 describes the operation of the health-insurance scheme and this is followed by a description of the data in Section 3. Section 4 discusses the empirical strategy, including the hypotheses, the methodology used, potential sources of bias, and some descriptive statistics. Section 5 presents the results of the analyses. Section 6 discusses the implications of our results, while Section 7 draws some conclusions. 2. MICROCARE’S HEALTH-INSURANCE SCHEME Microcare is a regulated and licensed private insurance company that provides health insurance in Uganda to employees in the formal sector and to groups of households in the socalled informal sector, such as self-employed farmers. The communities in the informal sector are previously established groups, such as credit groups or farmers’ associations, or groups that have organized themselves in order to access insurance. Microcare’s health-insurance schemes are concentrated in the rural areas around Kisiizi and Kisoro and in the urban centers of Kampala and Entebbe. This study is restricted to Kisiizi and Kampala. Around Kisiizi there are 74 groups, with a total of 3,134 households (approximately 15,512 people) in the scheme. In Kampala and Entebbe, 18 groups of households (approximately 1,396 individuals) are part of the health-insurance scheme. Participation rates in

Microcare’s informal-sector health-insurance scheme are approximately 15% in Kisiizi and less than 1% in Kampala. Microcare offers a health-insurance package covering outpatient and inpatient services but excluding medication for chronic illnesses such as HIV/AIDS, hypertension, and diabetes (McCord & Osinde, 2002). Microcare contracts public and private health-service providers to offer services to members insured in the schemes. Those in Kampala can choose between three health-service providers with user fees for Out Patient Department (OPD). In Kisiizi, only Kisiizi Hospital provides services to the insured and charges user fees for OPD and an admission fee for inpatient services. Premiums for informalsector members have to be paid as an annual lump sum, are non-refundable, and vary according to a household’s size and location. At the time of this study, the annual premium in Kampala and Entebbe was USh 149,000 (US$ 80.54) for a family of four in the scheme with additional premiums for extra members: USh 52,000 (US$ 28.11) for each extra adult and USh 26,000 (US$ 14.05) for an extra child. The Kisiizi scheme charged annual premiums starting at USh 24,000 (US$ 12.97) for a family of 4 while a family of 8 paid USh 32,000 (US$17.29) per annum and a family of up to 12 members paid USh 40,000 (US$ 21.62). On average, this amounts to between 1% and 2% of a household’s annual expenditure (UBOS, 2006). The premiums paid by informal-sector clients are not sufficient to cover all Microcare’s costs. These are cross-subsidized by the premiums paid by their clients in the formal sector. In Kampala, the microfinance institution Foundation for International Community Assistance (FINCA) provides loans to clients to pay their health-insurance premiums. In Kisiizi, such services were planned by Uganda Microfinance Limited (UML) but were not yet available to the respondents at the time of our study. 3. DATA The data for this study were collected in June/July 2006 and cover insured and uninsured households in and around Kisiizi and Kampala. 3 The sample for this study was a convenience sample; in both size and selection. It is not representative of the population in Kisiizi or Kampala and the uptake of insurance in our sample was higher than the true rates of insurance in these areas. Based on availability and group size, the researchers visited five community groups around Kisiizi that were part of Microcare’s insurance scheme: Kyondo, Muhanga/Butale (a combined group for survey purposes), Rwababishisha, Kamakinda, and Kamoobwa. Not all the members of these groups were insured. The data were collected using individual interviews with members during their weekly group meetings with Microcare. Such meetings, headed by a sales representative from Microcare, were used to discuss disease prevention and provide information about Microcare’s insurance products. At the end of these group meetings, every member, usually the head of the household, was asked to participate in a private interview with the researchers and a Ugandan interpreter. The questionnaire included questions about their health-seeking behavior, and their use of health insurance and other strategies to finance health care. In addition to these community group members, the researchers randomly selected people at the market place in Kisiizi town and patients waiting for treatment at Kisiizi hospital to participate in the same survey. In and around Kampala, the questionnaire was administered in the area where two microfinance institutions, UML

HEALTH INSURANCE AND OTHER RISK-COPING STRATEGIES IN UGANDA Table 1. Research sites and insurance take-up Number of observations

Percentage insured (%)

Rural Muhanqal/Butale Kamakinda Kamoobwa Rwababishisha Kisiizi shops Kisiizi hospital Kyondo Urban Kajjansi (UML) Kasanqati (UML) Nsambya railways (FINCA) Tklezimbe-Klabiqalo (FINCA)

183 28 32 16 27 15 50 15 76 19 32 16 9

77 100 78 94 100 27 60 73 14 11 60 44 0

Total

259

58

and FINCA, are working. In the FINCA area, the researchers used the same procedure followed for the Kisiizi groups. Participants in two FINCA credit groups, Nsambya Railways and Tklezimbe-Klabigalo, were invited to participate in the survey at weekly loan-repayment meetings when public-health information was given and the Microcare insurance product was discussed. Respondents in the UML area did not belong to a group and were selected at random. The respondents consisted of new clients enrolling in the credit scheme and clients with outstanding loans who had come to make a repayment. In total, 259 respondents, each representing a different household, were interviewed, 183 in and around Kisiizi and 76 in and around Kampala. Table 1 provides an overview of the research locations and the number of respondents interviewed at each location. The groups in Muhanga/Butale and Rwababishisha were made up of insured respondents only, while other rural groups had both insured and uninsured respondents. Some group members of the urban neighborhoods were insured, but none of the participants in Tklezimbe-Klabigalo was insured. Given the cross-sectional nature of our study and the nonrandom selection of the sample, the results presented here should be treated as indicative only. We cannot claim that the differences observed between the insured and uninsured are entirely due to having health insurance or not, but the observed differences do warrant further study. The potential biases related to the sample selection are discussed in more detail in the next section.

371

To test these hypotheses, we estimated the relationship between insurance and OOP health expenditure and the incidences and value of risk-coping strategies, respectively, using OLS and Probit regressions. Measuring the impact of health insurance on health expenditure and risk-coping strategies can, however, be problematic given the potential endogeneity of the health-insurance variable. This endogeneity is related to the possibility that a household’s decision to insure its members may be related to unobserved characteristics that also affect the use of health-care services and other risk-coping strategies. If this is the case, a negative relationship between health insurance and OOP expenditure (or risk-coping strategies) cannot be attributed to insurance but to underlying unobserved characteristics, thus resulting in an overestimation of the effects of insurance. In our study set-up, using Microcare group meetings and (potential) microfinance members as respondents, a source of bias was related to the selection of the sample. People who attend such meetings, or have loans, can have unobserved characteristics that make them different from non-members and these characteristics may influence both health-seeking behavior and insurance take-up. Assuming that these characteristics have a positive relationship with insurance (i.e., such people are more likely to be insured), if differences are found between those insured and uninsured in groups, these cannot be attributed to such unobservable characteristics. This means that in our analyses, the effect of insurance is likely to be underestimated vis-a`-vis un-insured households outside groups and without access to loans. As far as possible, we tried to control this potential bias by including variables on group membership and access to micro-finance loans. Following Waters (1999) and Jutting (2004), we also used the exogeneity test proposed by Smith and Blundell (1986) to assess whether ‘‘being insured” could be treated as an exogenous variable in our context. First, we estimated a reduced form of insurance and predicted the insurance variable based on these results. 4 We then included the predicted and the actual insurance variables as explanatory variables in the healthcare expenditure and coping-strategies regressions. A non-significant coefficient for the predicted insurance variable was interpreted as an indication that insurance could be treated as exogenous. Estimations reported in the appendix indicate that health insurance could be treated as an exogenous variable in four of the six analyses, where the predicted insurance variable does not enter the regression equations with a significant coefficient. In the analyses of OOP expenditures per illness and incidences of sales of assets, endogeneity of the insurance variable cannot be ruled out.

4. EMPIRICAL STRATEGY

(b) Estimation outline

(a) Hypotheses and potential sources of bias

In our analysis, we used six dependent variables. OOP health expenditures were measured as the total costs for treatment, including transport to the medical facility and the insurance premium (for those insured), in USh. These were measured at the household level and divided by the number of illnesses in a household, that is, per illness. The use of other risk-coping strategies was measured as a binary variable that equaled one if households sold assets (or borrowed money) to finance their OOP expenditures. The proceeds of the riskcoping strategies were measured per illness, as the amount obtained through the sale of assets (or borrowed), maximized at total OOP expenditures and divided by the number of illnesses. 5 Three sets of independent variables are included in the analysis. First, we controlled for health risks by including variables

This research project considered the relationship between health insurance and levels of OOP expenditures on health and the use of other coping strategies to finance health care. Microcare covers a large part of OOP expenditures for those with health insurance and, for this reason, we expected that OOP expenditures—with at least a minimum for transport and co-payment fees—would be lower for households with health insurance. Health insurance was also expected to influence the use of other risk-coping strategies, such as borrowing and the sale of assets to cover health-care expenses. We expected households with health insurance to be less likely to use such strategies to pay for health care and if they did, the amount generated through such strategies would be lower.

372

WORLD DEVELOPMENT

on the frequency of illnesses in the household, making a distinction between malaria, chronic illnesses, and other illnesses. The variables report the number of cases the household experienced over the past year. 6 Given their exclusion from the insurance policy and the potential recurrent need for medication, chronic illnesses were expected to be associated with higher OOP expenditures. Given the high frequency of malaria in the rainy season, cases were also expected to lead to higher OOP expenditures at household level but to lower OOP expenditures per illness, as the cost of treatment is relatively low. Second, we included four indicators related to a household’s ability to pay for health expenditures. Based on a ranking of estimated annual cash expenditures, we defined two dummy variables on wealth, one for rich households and one for poor. 7 Two other dummies were related to employment status: occupation equaled one if a household head was working in the formal sector, while unemployment equaled one if the head of household was neither formally employed nor working in business or in agriculture. We expected rich households and household heads with formal-sector employment to have higher OOP expenditures (because of a higher capacity to pay) and to use fewer other risk-coping strategies, while unemployed and poorer household heads would be more likely to have lower OOP expenditures and make greater use of other risk-coping strategies. As access to loans from micro-finance organizations and membership in informal insurance or micro-finance groups is likely to influence whether or not a household has health insurance, we included a variable loan that equaled one if the respondent was interviewed in a micro-finance community and a variable group that equaled one if the respondent was a member of an insurance group and interviewed after a group meeting. As emphasized earlier, membership of an insurance group does not necessarily mean that households are insured. Larger households are less likely to take out insurance given the higher costs involved for each additional member. For this reason, we also included household size. Finally, differences in the availability of and distance to health-care providers, the costs involved in seeking health care as well as other local characteristics also influence OOP expenditures and risk-cop-

ing behavior. These factors are controlled for by including community fixed effects and a dummy that equals one for households from rural areas. Table 2 provides descriptive statistics of the variables used in our analysis and distinguishes respondents with and without insurance. 8 There are a few interesting observations to note here. First, households with insurance reported fewer illnesses compared to uninsured households, 2.49–3.76 cases. Malaria was the most frequently reported illness: more than half of all illnesses in the past year were diagnosed as malaria, with uninsured households having a higher percentage of these cases. Second, we found that OOP expenditures on health care were significantly higher in the uninsured households: in the last 12 months they had spent USh 186,640 (US$ 100.88) on average, compared to the insured households who had spent USh 83,420 (US$ 45.09). OOP expenditures per illness show similar differences: USh 59,930 (US$ 32.39) for uninsured households compared to USh 31,160 (US$ 16.84) for insured households. If insurance premiums are excluded from the calculations (not reported in the table), OOP expenditures on health care are on average USh 102,320 (US$5 5.31). In that case, the difference in level of expenditure is even more pronounced, averaging USh 186,640 (US$ 100.89) for uninsured households compared to USh 42,020 (US$ 22.71) for insured households. The fewer illnesses reported by insured respondents and the lower costs per illness also suggest moral hazard (seeking care more often or seeking higher quality care because the respondent is insured) is not relevant in this context. The results also show that OOP expenditure was higher than the ability to pay in many cases: (only) 44% of the uninsured households and 56% of those insured had enough cash to pay for health care. To finance OOP expenditures, households used other risk-coping strategies. Borrowing to pay for health care or health insurance is somewhat less common among insured households and, if they do so, the amount borrowed is significantly lower. Insured households more often sold assets to pay for health care or health insurance than uninsured households but the amount for which insured households sold assets was significantly lower. On average, uninsured

Table 2. Descriptive statistics of key variables All (N = 259) Mean (SD) Illnesses Illness episodes per household Malaria Chronic Other

3.02 2.08 0.24 0.71

Household characteristics Household size Formal sector employment (Yes = l) Unemployment (Yes = l) Rich (Yes = l) Poor (Yes = l)

6.23 (2.69) 7% 4% 30% 35%

Health care expenditures Out-of-pocket expenditures including health insurance premium (1,000) Cost per illness (l,000) Able to pay OOP (Yes = l) Indebted to pay OOP (Yes = 1) Sold assets to pay OOP (Yes = l) Amount borrowed to pay OOP (1,000) Value of assets sold to pay OOP (l,000)

(2.46) (2.10) (0.53) (1.22)

Uninsured (N = 108) Mean (SD) 3.76 2.56 0.23 0.97

(2.97) (2.54) (0.49) (1.59)

Insured (N = 151) Mean (SD) 2.49 1.75 0.24 0.53

t-test difference In means

(0.39) (1.64) (0.56) (0.81)

0.0000 0.0021 0.9177 0.0038

6.53 (3.07) 13% 8% 30% 40%

6.01 (2.36) 2% 1% 29% 32%

0.1288 0.0310 0.0050 0.6560 0.4570

126.46 (221.41)

186.64 (305.74)

83.42 (114.74)

0.0002

43.16 (78.91) 131 (51%) 110 (43%) 67 (27%) 115.32 (189.59) 62.21 (117.49)

59.93 (112.42) 47 (44%) 52 (48%) 19 (20%) 183.67 (253.64) 138.94 (214.08)

31.16 (36.72) 84 (56%) 58 (38%) 48 (32%) 55.26 (65.37) 35.03 (17.63)

0.0036 0.0549 0.1189 0.0483 0.0067 0.0013

HEALTH INSURANCE AND OTHER RISK-COPING STRATEGIES IN UGANDA

households sold assets worth USh 138,940 (US$ 75.10) while insured households sold USh 35,030 (US$ 18.94) worth. The higher frequency of asset sales for insured members may be related to the area of residence of households. Selling off assets is, for example, a common strategy in rural areas but not in urban areas, while sources of credit (and the amount borrowed) also differed considerably between rural and urban areas, with more formal credit services available in urban areas. 5. ESTIMATION RESULTS In this section, we use regression analyses to explore the relationship between health insurance and OOP expenditures on health care, the incidence and the level of asset sales to finance health expenditure and the incidence and levels of borrowing to finance health expenditure. All analyses include community fixed effects and standard errors are corrected for clustering at community level. The coefficients for the fixed effects are not reported in the tables and in some analyses the rural, group, or loan dummy was dropped due to a lack of variation in these variables in the regression. First, we consider the relationship between health insurance and OOP health expenditures, measured at the household level and reported per illness episode, using an OLS regression (Table 3). In both cases, health insurance has a significant negative effect on the level of OOP expenditures, decreasing household OOP expenditures by roughly USh 55,000 (US$ 29.73), while OOP expenditures per illness are USh 31,252 (US$ 16.89) lower for insured households. As expected, the number of episodes of illness experienced increased the level of OOP expenditure at household level, especially for chronic illnesses and, to a lesser extent, for malaria and other illnesses. Malaria on the other hand, is negatively associated to OOP expenditures per illness, indicating that malaria is a ‘‘low-cost” illness. Disaggregated figures on treatment confirmed this: malaria treatment costs approximately USh 18,000 (US$ 9.73) in Kisiizi and USh 39,000 (US$ 21.08) in Kampala, while the average cost of illnesses was USh 43,000 (US$ 23.24) and USh 50,000 (US$ 27.03), respectively. OOP expenditures do not vary significantly with employment status, household size, or wealth status, while formal-sector workers have higher OOP expenditures, possibly due to the

373

accessibility and affordability of health care for this group. Households with access to micro-finance loans reported lower OOP expenditures per reported illness. The reported results are robust to the exclusion of the health-insurance premium from OOP expenditures. The analysis presented in Table 4 addresses the relationship between health insurance and asset sales. The first panel presents a probit regression on the incidence of asset sales and includes all respondents with OOP for health, while the second panel is an OLS regression on the average value of asset sales per illness and includes only respondents with asset sales. Insured households are less likely to engage in asset sales to finance health-care expenditures and on average spend less. The value of assets sold per illness is also lower when a household has experienced more cases of malaria. Households that borrowed to pay for either health care or health insurance were more likely to sell assets. The value of assets sold was however not significantly lower for households that borrowed money as well, possibly related to the indivisibility of assets. Household characteristics such as wealth, employment status, and household size do not explain the observed differences in the number of incidences or the level of asset sales. Table 5 shows the estimation results of borrowing money to pay for OOP expenditures. The first analysis was a probit on the incidence of borrowing money and included respondents with OOP expenditures. The OLS regression on the amount of money borrowed per illness case presented in the second panel only included respondents who borrowed money to pay for OOP expenditures. We found no evidence of households with health insurance being less likely to borrow money to meet OOP expenditures and this finding is robust to the exclusion of the insurance premium from OOP expenditures. When borrowing to pay for health expenditures, we did however find that households with health insurance borrowed less money per reported illness (a reduction of USh 42,828 or US$ 23.15). The type of illness is not associated with the incidence of borrowing but other illnesses reduce the amount of money borrowed per illness. Households that sold assets to pay medical bills were more likely to have a loan too. Job status also affected incidences of borrowing; both the unemployed and those with a job in the formal sector borrowed less often. Finally, households that were part of a group borrowed significantly more often to finance OOP and the value of the loan was significantly lower.

Table 3. Insurance and OOP expenditures (in 1000 Ush) OLS OOP expenditures

Health Insurance (1 = yes) Chronic illnesses Malaria cases Other illnesses Rich (1 = yes) Poor (1 = yes) Unemployed (1 = yes) Formal sector job (1 = yes) Household size Access to MF loans (1 = yes) Interviewed after group meeting (1 = yes) No. of observations

OLS OOP expenditures

HH level

t-stat

Per illness

t-stat

54.864*** 128.639*** 11.071** 22.777* 12.470 34.558 14.861 219.331*** 8.435 40.753 26.701

( 3.718) (3.745) (2.274) (2.053) (0.444) ( 0.588) ( 0.563) (3.733) (0.944) ( 1.450) ( 1.058)

31.252* 22.775 6.238** 3.813 4.501 11.763 14.521 99.771*** 3.636 20.162** 6.760

( 4.729) (1.254) ( 2.304) ( 0.962) (0.314) ( 0.480) ( 1.269) (3.309) (1.350) ( 2.673) ( 0.644)

259

Community fixed effects included, standard errors are corrected for clustering at the community level. * .1. ** .05. *** .01.

259

374

WORLD DEVELOPMENT Table 4. Health Insurance and asset sales Probit Sold assets

Health insurance (l = yes) Chronic illnesses Malaria cases Other illnesses Rich (l = yes) Poor (l = yes) Rural (l = yes) Formal sector job (l = yes) Household size Borrowed for OOP (l = yes) Interview after group meeting (l = yes) No. of observations

OLS Value sold

1 = yes

t-stat

0.565** 0.168 0.069 0.037 0.132 0.145 0.642 0.292 0.006 0.353* 0.169

( 2.012) ( 0.747) (1.290) ( 0.340) ( 0.529) ( 0.599) (1.478) ( 0.458) ( 0.139) (1.747) ( 0.388)

Per illness

t-stat

48.828*** 4.191 7.543** 2.511 12.485 13.783 6.628

( 2.160) (0.271) ( 2.093) ( 0.279) (0.731) (0.833) (0.220)

2.915 1.642 16.268

245

(0.972) (0.108) (0.454) 60

Community fixed effects included, standard errors are corrected for clustering at the community level. * .1. ** .05. *** .01.

Table 5. Health Insurance and Borrowing behaviour Probit Borrowed

Health insurance (l = yes) Chronic illnesses Malaria cases Other illnesses Rich (l = yes) Poor (l = yes) Rural (1 = yes) Unemployed (l = yes) Formal sector job (l = yes) Household size (l = yes) Sold assets for OOP (l = yes) Interviewed after group meeting (l = yes) No. of observations

OLS Value borrowed

l = yes

t-stat

Per illness

0.056 0.147 0.008 0.079 0.161 0.348 0.375 0.561* 0.559*** 0.017 0.352* 0.345***

(0.285) (0.880) (0.146) ( 1.051) ( 0.720) ( 1.011) ( 1.556) ( 1.816) ( 2.285) ( 0.469) (1.771) (2.718)

42.828* 11.295 1.380 7.756** 6.479 15.696 0.094 3.974 160.496 1.322 3.376 14.382**

245

t-stat ( ( ( ( ( (

2.030) 0.621) 0.872) 2.675) 0.875) 1.510) (0.006) (0.154) (1.439) (1.141) (0.718) ( 2.541)

146

Community fixed effects included, standard errors are corrected for clustering at the community level. * .01. ** .05. *** .01.

6. DISCUSSION The analyses presented in Section 5 aimed to provide an understanding of the relationship between health insurance and OOP health-care expenditures and other risk-coping strategies. When controlling for frequency of illness and household and location characteristics, we found that OOP expenditure on health care (including the health-insurance premium) was lower for insured households, both at the total level of expenditures and per illness. Insured households sold assets less frequently, the value of the assets they sold was lower and the amount borrowed per illness to pay for health care and health insurance was also lower. The first two results have been reported elsewhere in the literature on health insurance (Ekman, 2004; Jowett et al., 2003; Jutting, 2004; Preker et al., 2001; Ranson, 2002) but, to our knowledge, this is the first study

to quantitatively assess the relationship between insurance and other risk-coping strategies. Health insurance is a risk-coping strategy in itself and is unlikely to make other risk-coping strategies redundant for at least two reasons. First, insurance policies cover a limited number of costs. They exclude, for example, medication for chronic illnesses which results in insured households still having high and sometimes unpredictable costs. On average, insured members in the rural areas paid over USh 54,000 (US$ 29.19) per chronic illness. Insured members in urban areas pay over USh 108,900 (US$ 58.86). In rural areas this amount is only slightly lower than the amount uninsured members pay, while in urban areas uninsured households pay almost Ush 140,000 (US$75.67). Additionally, health-care providers ask user fees for OPD visits and admissions from both the insured and uninsured. An OPD visit to Kisiizi

HEALTH INSURANCE AND OTHER RISK-COPING STRATEGIES IN UGANDA

Hospital was USh 1,000 (US$ 0.54) for the insured and USh 1,200 (US$ 0.65) for the uninsured, while admissions cost USh 5,000 (US$ 2.70) for everyone. An OPD visit to the health-care facilities in Kampala varied between USh 2,500 (US$ 1.35) and USh 5,000 (US$ 2.70). In addition, transport costs can be high in the case of severe or repeated illnesses or if it is a long way to the nearest health-care facility, as is often the case in rural areas. Transport costs in and around Kisiizi amounted to about 31% of total OOP expenditures on medical care, ranging, for example, from 53% in Kyondo to close to zero in Kamoobwa and at Kisiizi shops. In urban neighborhoods, households allocated on average 18% of their total OOP expenditure on health care to transport costs. Such additional costs are common in small-scale insurance schemes. Apart from the fact that co-payments are necessary to prevent moral hazard, the Microcare scheme (and many other micro-insurance schemes) provide cover for a limited group of people who cannot afford to pay high premiums. The limited risk pool and low premiums make it necessary to exclude the aforementioned costs. Although relatively low, expenditures for co-payments and transport may still force households to borrow money or sell assets as many households are cash constrained, especially in rural areas. Schemes should be careful not to charge co-payments beyond the level needed to prevent moral hazard. The dissimilar relations between insurance and asset sales compared to insurance and borrowing can be understood in relation to the indivisibility of assets. Although there is some variation in the value of assets, such as a goat and a cow, assets are generally indivisible and have to be sold in their totality, even if only a fraction of the value is required to meet cash needs. The value of loans on the other hand is more continuous and can be more responsive to the amount of cash required at any one time. A second reason why other risk-coping strategies have not been made redundant by health insurance is that numerous respondents reported difficulties in paying their health-insurance premium and borrowed money or sold assets to do so. Information from the survey shows that only 37% of insured household heads were able to pay their premiums and more than half of those insured (55%) borrowed money to pay them. Among the insured respondents who borrowed money, over 82% borrowed from other members in their community. Others, especially in urban areas, used a micro-credit loan to pay the premium. FINCA Uganda facilitates the payment of the premium by deducting it from the group loan. The groups within Nsambya Railways and Tklezimbe-Klabigalo had access to this type of loan. These findings suggest that more tailor-made premium-payment schedules be recommended, as connection to a micro-finance institution would facilitate the payment of premiums and encourage saving. The above discussion confirms what has been advanced before, for example, by Jutting (2005), that health insurance cannot offer households full protection against the financial effects of health risks. Health insurance can at best be an integral part of a wider risk-management strategy containing a number of formal and informal instruments. In this discussion, it should also be remembered that our results are based on just one case study that was limited to

375

Microcare’s insurance schemes in Kampala and Kisiizi and these may not be representative of all members or across Uganda. They must also be interpreted within the context of the Microcare’s scheme in terms of size, outreach, structure, benefit package, and cost. Other insurance schemes or risk-coping programs may have different characteristics and possibly different relationships with other risk-coping strategies. In any future work, information should be collected on the incidence of risk-coping strategies to gain more indepth information on the level of sales or loans involved and, particularly, a household’s remaining assets and outstanding debts. 7. CONCLUSION Health-insurance schemes have been advanced over the past decades as an appropriate instrument in reducing the impact of health risks in developing countries and in poverty-reduction strategies in general. Protection against high, unexpected medical costs through health insurance is hypothesized to reduce OOP health expenditures and the use of other (costly) risk-coping strategies such as borrowing money and the sale of assets. Although several empirical studies have indeed confirmed a reduction in health expenditures, the hypothesized change in risk-coping behavior still remained unsubstantiated. Based on a survey in five rural and two urban communities in Uganda, this study has addressed the relationship between insurance and OOP health expenditures and the use and proceeds of other risk-coping strategies. In line with the hypothesis and empirical results obtained in other contexts, we found that insurance reduces OOP expenditures on health care (including the health-insurance premium). We also showed that insured households sell assets less frequently and, when they do, the value of the assets sold per illness is lower than the sums achieved by uninsured households. At the same time, insured and uninsured households borrow money equally, but those insured borrowed less money per illness compared to uninsured households, potentially influencing a household’s level of indebtedness. The findings presented here imply that health insurance cannot offer households full protection against the direct financial effects of health risks. The availability of cash to meet, for example, transport costs, user fees, exclusions from the health package or the health-insurance premium, remains problematic and is an important reason for household borrowing (or the selling of assets). At the same time, finding a lower incidence of asset sales for insured respondents tentatively suggests that health insurance can make a positive contribution to reducing other risk-coping strategies and thereby to improving a household’s well-being more generally. Given the potential sources of bias in our sample, this study should be replicated in a representative sample and different contexts to ascertain whether this relationship can indeed be attributed to health insurance and how the results are affected by the setup and context of Microcare’s schemes.

NOTES 1. The literature on health financing also stresses the importance of insurance as a means of improving the quality of health-care services that are largely financed through patients’ OOP expenditures. See, for example, Wiesmann and Jutting (2001), Preker, Carrin, Dror, and Jakab (2001),

Preker et al. (2002) and Ekman (2004). In this paper, we consider insurance only from the perspective of households and do not address issues related to health-system financing.

376

WORLD DEVELOPMENT

2. Scheil-Adlung et al. (2006) have reviewed health-insurance schemes in Kenya, Senegal, and South Africa. In Kenya and Senegal they found that insured households sold assets less frequently, while the insured in South Africa and Senegal reported lower incidences of borrowing. It is unclear whether the differences between the insured and uninsured are statistically significant and whether they are robust when controlling for household characteristics. 3. For a more detailed report on this study, see Wilms (2006) and Verweij (2007). 4. The reduced form could only be estimated in the communities with sufficient variation in independent variables and is therefore restricted to 195 of the 259 observations. This may limit the validity of the test as some communities are excluded from the analysis. 5. In 10% of cases, households borrowed more money or generated more money from the sale of assets than was needed to meet OOP expenditures. This is likely to be related to the indivisibility of assets such as goats or cattle. For loans, it is not clear why more money is borrowed than is needed to pay OOP expenditures but this could be related to the fact that patients borrow money before they seek treatment and the actual costs incurred are lower than were originally expected or they budget to allow for a safety margin.

6. The one-year recall period may cause recall bias but does allows us to capture the incidences of seasonal-related illnesses, such as malaria. 7. No detailed information on expenditures is available from the survey. However, respondents were asked to estimate their levels of cash expenditure in an expensive month (when school fees were due) and in a normal month. With these estimates, annual cash expenditures were calculated as the sum of three expensive months and nine normal months. Compared to Uganda National Household Survey expenditure levels, estimated expenditures are roughly similar in urban households, while the estimated cash expenditure in rural households represents approximately 75% of the total expenditure in the rural areas (UBOS, 2006). Based on our measurement of cash expenditures, households were ranked and assigned to terziles: poor, average, rich. This is a very crude measure of a person’s ability to pay as it only captures cash expenditures, which are usually higher for urban households than those in the rural areas. For this reason, we constructed area-specific terziles, for rural and urban households separately and ranked the two types of households accordingly. 8. As the insured and uninsured in the sample are not randomly and proportionally selected from the insured and uninsured population in Kampala and Kisiizi, the differences presented in Table 2 cannot be generalized to the total population of insured and uninsured in these locations, nor to Uganda as a whole.

REFERENCES Asenso-Okyere, W. K., Anum, A., Osei-Akoto, I., & Adukonu, A. (1998). Cost recovery in Ghana: Are there changes in health care seeking behaviour?. Health Policy and Planning, 13(2), 181–188. Atim, C., & Sock, M. (2000). An external evaluation of the Nkoranza community financing health insurance scheme, Ghana. Technical Report No. 50. Bethesda, MD: Partnerships for Health Reform Project, Abt Associates Inc. Bennett, S., Creese, A., & Monash, R. (1998). Health insurance schemes for people outside formal sector employment. ARA Paper No. 16, WHO, Geneva. Bogale, T., Mariamand, D. H., & Ali, A. (2005). Cost of illness and coping strategies in a coffee-growing rural district of Ethiopia. Journal of Health Population and Nutrition, 23(2), 192–199. Dekker, M. (2004). Sustainability and resourcefulness. Support networks in times of stress. World Development, 32(10), 1735–1752. Dercon, S., Hoddinott, J., & Woldehanna, T. (2005). Vulnerability and shocks in 15 Ethiopian villages, 1999–2004. Journal of African Economies, 14(4), 559–585. Dror, D. M., Soriano, S., Lorenzo, M. E., Sarol, J. N., Jr., Azcuna, R. S., & Koren, R. (2005). Field based evidence of enhanced healthcare utilization among persons insured by micro health insurance units in the Philippines. Health Policy, 73(3), 263–271. Ekman, B. (2004). Community-based health insurance in low-income countries: A systematic review of the evidence. Health Policy and Planning, 19(5), 249–270. IFC (2009). The business of health in Africa. Partnering with the private sector to improve people’s lives. Washington, DC: International Finance Corporation. ILO (2008). Social health protection. An ILO strategy towards universal access to health care. Geneva: International Labour Organisation. Jowett, M., Contoyannis, P., & Vinh, N. D. (2003). The impact of public voluntary health insurance on private health expenditures in Vietnam. Social Science and Medicine, 56(2), 333–342. Jutting, J. (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. Jutting, J. (2005). Health insurance for the poor in developing countries. Burlington: Ashgate Publishers. Krishna, A., Lumonya, D., Markiewicz, M., Kafuko, A., Wegoye, J., & Mugumya, F. (2006). Escaping poverty and becoming poor in 36 villages of central and western Uganda. Journal of Development Studies, 42(2), 346–370.

Leliveld, A. (2006). Poverty trap or safety net? Dynamics in social security arrangements in Ugandan rural economies. Paper presented at the Conference ‘‘The End of Poverty in Africa?” African Studies Centre, Leiden, 16–17 March 2006. McCord, M. J. (2000). Health Care MicroInsurance: A synthesis of case studies from four health care financing programs in Uganda, Tanzania, India, and Cambodia, Micro-Save, Kenya. Nairobi: Microsave-Africa . McCord, M. J., & Osinde, S. (2002). Microcare Ltd. Health plan: Notes from a visit 17–21 June 2002. Kampala: MicroSave-Africa . Preker, A., Carrin, G., Dror, D. M. & Jakab, M. (2001). Health care financing for rural and low-income populations: The role of communities in resource mobilization and risk sharing a synthesis report. Prepared for the Commission on Macroeconomics and Health, World Bank Health, Nutrition and Population Discussion Paper. Washington, DC: The World Bank. Preker, A. S., Carrin, G., Dror, D. M., Jakab, M., Hsiao, W., & ArhinTenkorang, D. (2002). Effectiveness of community health financing in meeting the cost of illness. Bulletin of the World Health Organization, 80(2), 143–150. Ranson, K. (2002). Reduction of catastrophic health care expenditure by a community-based health insurance scheme in Gujarat, India: Current experiences and challenges. Bulletin of the World Health Organization, 80(8), 613–621. Scheil-Adlung, X., Carrin, G., Ju¨tting, J., & Xu, K. (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. Extension of Social Security, ESS Paper No. 24. Geneva: International Labour Organisation. Schneider, P., & Hanson, K. (2006). Horizontal equity in utilization of care and fairness of health financing: A comparison of micro-health insurance and user fees in Rwanda. Health Economics, 15(1), 19–31. Smith, R. J., & Blundell, R. W. (1986). An exogeneity test for a simultaneous equation Tobit model with an application to labor supply. Econometrica, 54(3), 679–685. UBOS (2006). Uganda National Household Survey 2005/2006. Kampala: Uganda Bureau of Statistics. Verweij, M. (2007). Influencing factors on the micro health insurance purchasing decision: Evidence from Uganda. Masters Thesis. Unpublished master thesis. University of Utrecht, Utrecht.

HEALTH INSURANCE AND OTHER RISK-COPING STRATEGIES IN UGANDA Waters, H. R. (1999). Measuring the impact of health insurance with a correction for selection bias – A case study of Ecuador. Health Economics, 8, 473–483. Wiesmann, D., & Jutting, J. P. (2001). Viable health insurance schemes in rural Sub-Sahara Africa. Quarterly Journal of International Agriculture, 50(4), 361–378. WHO. (2006). Reference Health financing: A strategy for the African region. Report of the Regional Director, AFR/RC56/10 17 June 2006, Regional Committee for Africa. Wilms, A. (2006) The financial impact of formal health insurance schemes: Evidence from Uganda. Unpublished Master Thesis. Vrije Universiteit Amsterdam, Amsterdam . Young, P., Mukwana, P., & Kiyaga, E. (2006) Microinsurance exploring ways to assess its impact. Washington, DC: Microfinance Opportunities/IRIS.

APPENDIX A See Tables A.1 and A.2.

377

Table A.1. Reduced form probit estimation of health insurance participation Probit Health insurance

Chronic illnesses Malaria cases Other illnesses Rich (=1) Poor (=1) Unemployed (=1) Formal sector job (=1) Household size lnterviewed in group meeting (=1) Access to MF loans (=1) No. of observations

l = yes

t-stat

0.049 0.070 0.080 0.181 0.438 0.199 0.192 0.094* 1.371*** 1.508*** 195

(0.239) ( 1.642) ( 1.295) ( 0.587) ( 0.960) (0.639) (0.491) ( 1.684) (13.177) ( 6.522)

Community fixed effects included, standard errors are corrected for clustering at the community level. Note: some communities are dropped from the analysis due to limited variation. * .1. *** .01.

(See Overleaf )

378

Table A.2. Regression result exogeneity test for participation in health insurance OLS OOP

Probit Borrowed

Probit Sold assets

HH level

t-stat

Illness

t-stat

1 = yes

t-stat

54.621*** 142.843 145.757*** 7.833 23.360 5.655 63.562 17.575 215.232** 4.995 29.889

( 3.840) ( 0.358) (4.366) (1.729) (1.229) ( 0.147) ( 0.660) ( 0.587) (3.315) (0.279) (0.189)

31.269*** 113.426* 29.521 8.581* 4.949 7.516 30.585 16.371 97.381** 0.512 40.682

( 5.571) ( 2.125) (1.390) ( 2.010) ( 0.903) ( 0.413) ( 1.048) ( 1.338) (3.135) (0.121) (1.167)

0.179 1.598 0.267 0.003 0.052 0.106 0.076 0.566** 0.565*** 0.043 1.446 0.412

(0.986) ( 0.664) (1.377) (0.042) ( 0.580) ( 0.312) ( 0.137) ( 2.182) ( 2.641) ( 0.453) (1.627) ( 0.401)

17.548

( 0.099)

33.569

(1.546) 0.692***

(3.551)

195

195

181

1 = yes

OLS Borrowed t-stat

0.710** 2.910* 0.221*** 0.121*** 0.049 0.354 0.292

( 2.407) (1.921) ( 2.957) (4.646) ( 0.367) ( 1.277) ( 0.708)

0.266 0.056 0.129

( 0.343) (0.605) ( 0.361)

0.038

( 0.040)

0.659*** 155

Community fixed effects included, standard errors are corrected for clustering at the community level. Note: some communities are dropped from the analysis due to limited variation. * .1. ** .05. *** .01.

Available online at www.sciencedirect.com

Amount

OLS Sold assets t-stat

Amount

t-stat

41.026 175.016 10.229 6.630** 12.316** 6.234 34.101 7.756 167.569 1.854 45.171 140.355*

( 1.893) ( 1.623) ( 0.491) ( 2.481) ( 2.982) ( 0.563) ( 1.848) (0.262) (1.468) ( 0.617) (1.316) ( 2.378)

49.121 128.534 2.961 7.766 4.153 40.081 34.949

( 1.018) (0.407) (0.085) ( 1.307) (0.914) (0.812) (0.658)

9.171 22.303

(1.475) (0.383)

7.989

(1.198) 17.362 36

(0.990)

(4.035) 102

WORLD DEVELOPMENT

Health insurance (1 = yes) Predicted Healt Insurance Chronic illnesses Malaria cases Other illness Rich (1 = yes) Poor (1 = yes) Unemployed (1 = yes) Formal Sector job (1 = yes) Household size Interview after group meeting Access to MF loans Rural (1 = yes) Sold assets (1 = yes) Borrowed (1 = yes) No. of observations

OLS OOP