Population Research and Policy Review 17: 55–70, 1998. c 1998 Kluwer Academic Publishers. Printed in the Netherlands.
55
Financing rural health services in Kenya GERMANO MWABU & JOSEPH WANG’OMBE University of Nairobi, Nairobi, Kenya
Abstract. The paper analyzes household expenditure on medical care and the willingness to contribute towards service improvements at government health facilities. The analysis is based on survey data from two rural districts in Kenya situated approximately four hundred miles apart. The main finding is that medical care expenditure rises as household income increases, but the probability of willingness to pay fees for service improvement at government clinics declines with income. Income is an important determinant of the willingness to participate in a hypothetical government insurance scheme, with the probability of participation falling as income rises. These results should be interpreted with caution because of the potential for incorrect reporting of the willingness to pay for services that have an element of a social good. The policy implications of the results are briefly discussed. Key words: Kenya, Medical care, User charges, Willingness to pay
1. Introduction A system of user charges for health services is increasingly being adopted in sub-Saharan Africa both to diversify sources of funds for health ministries and to promote efficiency in service provision and use. Some 29 African countries for instance, already have national systems of user fees for health care (Shaw & Griffin 1995). It appears likely that other countries in subSaharan Africa will soon experiment with this mechanism of health services financing because of severe budgetary constraints that governments in the region face in providing basic health services to the population (World Bank 1994). Thus, given the potential for widespread adoption of user charges in low-income areas, it is important to understand their budgetary effects as well as their probable demographic and welfare consequences. Budgetary effects of user fees in the health sector are straightfoward to compute and are relatively well documented in the health finance literature in Africa. During the period 1981 through 1986 for example, the proportion of revenue from user fees, as a percent of government recurrent expenditures on health services in some 16 African countries, varied from 0.5% in Burkina Faso to about 20% in Ethiopia (Shaw & Griffin 1995). Despite some notable cases, budgetary impacts of user fees in sub-Saharan Africa have
56
GERMANO MWABU & JOSEPH WANG’OMBE
been relatively minor. The welfare and demographic consequences of user fees however are largely unknown. By raising the cost of health services in government health facilities, user fees might motivate households to use alternative and more cost-effective forms of medical care, thereby saving resources that can be used to increase consumption of other goods and services, which in addition to medical care enhance household welfare. However, the fees might also prevent some patients from seeking medical treatments, in which case, the average consumption of health services in the population could fall, with deleterious consequences on social well-being. Thus, the welfare effects of user charges are ambiguous to start with. Except for a few cases (Gertler & Van der Gaag 1990; Dow 1995) attempts to measure these effects in African countries have been rare. Also rare, but slowly emerging, are studies that examine potential demographic effects of user charges for health care. These effects arise from the fact that even when fees are restricted to curative services, they could indirectly affect utilization of preventive services such as the maternal and child health services, and thus influence the demographic structure of the population. Recent studies in Niger (Yazbeck & Leighton 1995 and in Burkina Faso (Barlow & Diop 1995) report mixed effects of user charges for curative care on utilization of preventive health services. In Niger, both the number of visits for chargeable outpatient care and the number of visits for free prenatal services increased in areas where revenue from fees was used to improve the quality of curative services. In view of this complementarity, a user fee on curative services should reduce demand for preventive services whenever it lowers demand for outpatient care. Furthermore, the theory of consumer behavior predicts that other things being equal a user charge on preventive services would reduce the rate at which they are used. As is obvious, this prediction rests on a very stringent assumption that other things remain unchanged as the user fee is implemented. In Burkina Faso, a user fee on prenatal services stimulated greater enrollment in prenatal care programs because the fee was associated with an improvement in service quality (Barlow & Diop 1995). Thus through their effects on preventive care, user charges on curative care can have important demographic outcomes, such as changes in fertility and mortality rates. These changes might eventually alter the age structure of a population (see Navaneetham, 1993 for long-term effects of food prices on adult mortality rates in India). The effects of user fees on the population might thus be broader than is commonly supposed. The aim of this paper is not to assess the economic and demographic effects of user charges for curative services, but to provide information on determinants of the willingness to pay for such services. The factors influencing household expenditures on private medical care are also examined, as well as the willingness to contribute towards a medical insurance scheme.
FINANCING RURAL HEALTH SERVICES IN KENYA
57
Information on variations in private medical expenditures according to some household characteristics, for instance income and age, might be relevant in anticipating expenditure patterns on public health services before the introduction of fees. Such information can be used to avoid adverse effects of fees. The need to examine medical insurance along with user charges arises from the fact that insurance enables people to pool risks of inability to afford medical care. The remainder of the paper is organized as follows. Section 2 describes the institutional context in which Kenya’s health services are provided; Section 3 describes the study areas, the sample selection procedures and the data. Section 4 outlines an analytic framework used to analyze the data and to interpret the results. The results are presented and discussed in Sections 5 and 6 respectively. Section 7 concludes with policy implications of the study. 2. Institutional setting The bulk of curative and preventive health services in Kenya are provided by government and church organizations (Republic of Kenya 1989b). In addition to these institutions, there exists a wide range of private and nongovernment hospitals and health centers in urban and rural areas. Health facilities at workplaces, informally known as ‘company clinics’ are an important feature of medical care provision in industrial areas of major towns. Traditional healers in rural as well as in urban areas are a key factor in the overall Kenyan health care system, but as is shown in Section 5 (Table 2) only a small fraction of the population in study areas reported having used them. Another notable characteristic of the modern health care system is the existence of pharmacies, solo medical practitioners and numerous retail outlets for nonprescription drugs throughout the country. Briefly, there is a strong private medical care sector in Kenya. Even so, the bulk of the population, particularly in the sample analyzed here, relies heavily on the public health sector for medical care. In the private subsector, the services are provided at a fee, with the fee for preventive services in some missionary health facilities being subsidized by the government. Before 1989, curative and preventive services in government health facilities were provided to the public free of charge, but after that date, people had to pay for curative services (Republic of Kenya 1989a; Collins et al. 1996). However, service provision in government dispensaries continued to be free of charge. Given the external pressure for market-based reforms in the health sector and in the rest of the economy (1993, 1994) it is likely that in the near future user charges will be increased in government hospitals and health centers. In the not too distant future, fees are also likely to be introduced for the first time in government dispensaries. The results reported
58
GERMANO MWABU & JOSEPH WANG’OMBE
in this paper are useful in assessing the feasibility of increasing or introducing fees in government health facilities at all levels of the national health care system. 3. Study areas and data The data were collected via a household survey from Kwale and Kirinyaga districts (in coast and central provinces), over a four month period from November 1988 to February 1989. Kwale district has two distinct settlements, one on the coastline and the other in the hinterland. The coastline settlement is dense, and has a well developed social infrastructure. Coconuts and cashew nuts are the main sources of income for the households on the coastline. The hinterland settlement in contrast is sparse and is characterized by mixed farming, supported by a thin social infrastructure, consisting mainly of seasonal roads and a few government offices. Kirinyaga district is a high potential agricultural region, with tea and coffee as the main cash crops. The main food crops in the area include maize, rice, and pulses. The district is the site of the largest irrigation scheme in the country, covering some 1200 hectares. The samples from the two districts are probability samples in the sense that they were picked from the general population at random. The sample from each district comprised at least 2 per cent of all the households in the district. Both Kirinyaga and Kwale districts had a combined population of over 775,000 people and some 97,000 households. We used a combination of multistage and cluster sampling methods to select the households for interviews. The first stage involved selection of primary sampling units, which happened to be the smallest administrative units in each district. Each district is divided into administrative areas known as locations, which in turn are split into sub-locations. At the time of the survey, Kirinyaga district had 16 locations and 76 sub-locations while Kwale had 21 locations and 62 sublocations. Two sublocations, primary sampling units, were selected randomly from each location using random numbers. Thus in Kirinyaga district, 32 sampling units were selected, whereas in Kwale 42 units were chosen. At the second sampling stage, two villages were selected using random numbers from the list of villages in the primary sampling units. Each village so selected was designated as a sampling cluster. In Kirinyaga district, 64 clusters were chosen, with the corresponding number in Kwale being 84. The third sampling stage involved selection of interviewees from a cluster of households or villages. Every household head in a village was to be interviewed if available. Interviews were conducted using a structured questionnaire, administered by enumerators from study areas. Interview sessions started at the chief’s home and progressed to the nearest home until all households in the cluster
FINANCING RURAL HEALTH SERVICES IN KENYA
59
were enumerated and interviewed. If a household head was not available for interview, the questionnaire was administered to an adult member in the household. The sample size in each district was determined by sizes of village clusters. The sample sizes were 1548 and 1518 households in Kirinyaga and Kwale respectively. In addition to these samples, another sample was drawn from Kwale district to study utilization of maternal and child health services. The sample consisted of 502 households, and was selected in the same manner as the main samples. The field sample sizes differ slightly from the analytic samples shown in Section 5 (Tables 2–5) because some questionnaires were incorrectly filled and others were spoilt. This resulted in missing values for certain variables, particularly the household income. All cases with missing values were eliminated from analytic samples. To the extent that the eliminated cases were random occurrences in the field samples, they had no effect on the magnitudes of the descriptive and regression statistics reported in Tables 2–5 in results section. The analytic sample sizes shown in Table 5 differ from those reported in Tables 2–4 because in some cases information on the willingness to contribute to an insurance scheme was missing, and again such cases were excluded from the samples analyzed. In both of the study areas, data were collected on household income, private medical care expenditures, accessibility to health facilities, health service utilization, and willingness to pay fees for government health services. The household heads were also asked whether they would be willing to contribute regularly towards a medical insurance scheme, if such a scheme were to be initiated by the government. In asking questions about willingness to pay user fees or to contribute to an insurance scheme, care was taken to ensure that respondents understood that the revenue raised would be used to improve the quality of government health services.
4. Model specification To quantify behavioral effects of the determinants of medical care expenditures at the household level, a linear expenditure model was estimated using ordinary least squares. Specifically, monthly private medical expenditures were regressed on household characteristics and on other relevant covariates such as distance to health facilities, perceived quality of medical care and malaria prevalence in the population. To identify the overriding factors in people’s willingness to pay fees for medical care or to contribute to an insurance scheme, conditional likelihoods of accepting to pay user charges or of agreeing to contribute to an insurance scheme were estimated using numerical methods (see, e.g., Greene 1993).
60
GERMANO MWABU & JOSEPH WANG’OMBE
Underlying the empirical equations that were estimated is a structural model of household behavior, the full exposition of which is beyond the scope of this paper (see, e.g., Deaton & Muellbauer 1980). However, a structural relationship underlying the reported willingness to pay fees or to contribute to an insurance scheme is sketched below. Following the literature on discrete choice econometrics (see, e.g., Akin et al. 1985, 1986; Greene 1993), let the benefit expected from acceptance or nonacceptance to pay fees be denoted as
Uij
=Z
ij
+ "ij ;
(1)
where Uij is the net benefit that household i expects from making a binary decision j (j = 1; 2); Zij is a vector of attributes that characterize household i and her decision options j ; is a vector of parameters to be estimated and "ij is a random disturbance of the net benefit that household i expects to get from the decision option j . In the above decision situation (acceptance or nonacceptance to pay fees), the household is assumed to decide in favor of an option that has a higher net benefit. A household’s acceptance or nonacceptance to contribute towards a medical insurance scheme is analyzed in exactly the same way as in Equation (1). The empirical problem in Equation (1) is to determine values of given data on Z from the household survey. If the values for are known, Equation (1) provides the formula for determining the benefits that households expect to get from the particular choices they make. This formula facilitates prediction of household choices given the decision rule (i.e., choose the most beneficial option) and given the sample data on Z . However, since the benefits (Uij s) from the decision options are not observable or measurable, it is not possible to estimate the values of the vector, , from Equation (1). The parameter vector, , can however be estimated by maximizing the likelihood of observing sample data, Z . In logarithmic form, the likelihood function for observing the sample data can be written as
L = i j Qij log Pij ;
(2)
with
Pij
= exp(U )=[exp(U ) + exp(U )] ij
ij
ik
(3)
L is the logarithm of the likelihood function (see, e.g., Greene 1991); Qij is equal to 1 if household i decided in favor of decision option j (for j not equal to k ) and equal to zero otherwise. And as before, Uij = Zij + "ij . It should be noted that the estimated values for , show how household characteristics affect the benefits from decision option j , and in turn, after a simple transformation, how the probability of a decision in favor of option j
FINANCING RURAL HEALTH SERVICES IN KENYA
61
Table 1. Description of variables A. Dependent variables EXPENDITURE Total household expenditure in Kenya Shillings on health services one month prior to the household survey. CHARGEOP_WI Willingness to pay for government health services; the variable takes a value of 1 if the household was willing to pay for outpatient services in government clinics and a value of zero otherwise. WILCOTRB_IN Willingness to contribute to an insurance program; this variable takes a value of 1 if the household was willing to contribute to a medical insurance scheme and a value of zero otherwise. B. Explanatory variables MONEY_PRICE Monetary cost of medical care in Kenya Shillings. DISTANCE_KM Distance in kilometers to the facility where the household sought or could have sought treatment. ADULTS_HLD The number of adults in the household. QUALITY_PE Perceived quality of care; a dummy variable with a value of 1 if quality of service at a health facility was perceived to be of high quality and a value of zero otherwise. MALARIA_PR Prevalence of malaria in the household; a dummy that takes a value of 1 for households reporting malaria as the most common health problem in the village, and a value of zero for households reporting otherwise. TRAVEL_TIME Time in minutes taken to travel to a health facility. WAIT_TIME Time in minutes spent at the health facility waiting for medical treatment. SOURCE_WHERE Source of treatment; the variable takes a value of 1 if the household sought treatment from a government health facility and a value of zero otherwise. INCOME_MO Monthly household income in Kenya Shillings.
is affected. Thus estimation of enables policy makers to assess the ex-ante reaction of the public to user fees, and to other policy intentions. Table 1 provides a description of variables that appear in the estimated equations.
5. Empirical results The results are presented in Tables 2–5, with descriptive statistics in Table 2, and the regression statistics in Tables 3–5, followed by an elaboration of both sets of results. Taking up the summary statistics first, it can be seen from Table 2 that the majority of the households in the two districts (70 percent) had sought
62
GERMANO MWABU & JOSEPH WANG’OMBE
Table 2. Sources of medical care in study areas Type of source
Percent of patients Kwale Kirinyaga
Government health facility Shop (over-the-counter medicines) Private medical facility Traditional healer Mission facility Self (home remedies) Chemist (with a doctor’s prescription) Religious or spiritual healer Other Total Sample size
69.3 17.6 3.8 3.3 2.6 1.6 0.6 0.3 0.9
71.0 12.2 8.6 0.7 5.4 0.1 1.4 0.1 0.5
100.0 1367
100.0 1347
Table 3. Ordinary least squares results [Dependent variable: Household expenditure (EXPENDITURE) on medical services one month prior to the household survey] Explanatory variables CONSTANT QUALITY_PE INCOME_MO MALARIA_PR DISTANCE_KM WAIT_TIME Adjusted R-squared Sample size a
Estimated coefficientsa Kirinyaga Kwale
Both areas
0.816 (4.94) 0.102 (1.90) 0.076 (2.36) 0.231 (0.98) 0.132 (1.52) 0.026 (0.46)
0.694 (3.94) 2.008 (0.04) 0.101 (3.18) 4.000 (1.58) 0.104 (1.14) 0.479 (0.75)
0.744 (6.24) 1.629 (1.64) 0.102 (4.57) 0.068 (2.64) 0.016 (0.28) 0.011 (0.26)
0.0057 1367
0.0048 1347
0.0085 2714
Absolute t-values in parentheses.
Statistically significant at 10 percent level. Statistically significant at 5 percent level.
medical treatments from government health facilities two weeks prior to the household survey. Further, additional data from the survey (same sample sizes as in Table 2) indicate a high level of self-reported morbidity in the
63
FINANCING RURAL HEALTH SERVICES IN KENYA
Table 4. Maximum likelihood estimation results [Dependent variable: Willingness to pay for government health services (CHARGEOP_WI)] Explanatory variables CONSTANT INCOME_MO MALARIA_PR MONEY_PRICE ADULTS_HLD WAIT_TIME Sample size Log-likelihood a
Estimated coefficientsa Kirinyaga Kwale 1.227
0.895
(6.99) 0.001 (3.31) 0.009 (0.87) 0.001 (0.97) 0.262 (5.02) 0.004 (1.350)
(4.308) 0.004 (0.99) 0.954 (5.80) 0.001 (0.62) 0.028 (0.62) 0.001 (0.29)
1367 871.7
1347 906.1
Both areas 0.489 (4.03) 0.001 (3.50) 0.609 (0.73) 0.001 (0.65) 0.135 (4.07) 0.146 (0.52) 2714 1855.6
Absolute t-values in parentheses.
Statistically significant at 5 percent level.
population: about 62 and 68 percent of households in Kwale and Kirinyaga districts respectively reported at least one episode of illness a fortnight prior to the interviews. Over the two weeks before the survey, 59 and 75 percent of the households reporting illnesses in Kwale and Kirinyaga districts had sought medical treatments, mainly from the closest health facilities. Government health facilities were the normal sources of medical care for most households (94–96 percent) in both districts. In addition to curative services, government and missionary facilities (Table 2) provide preventive health services, with a strong bias towards prenatal and immunization services. Provision of these services is also an important objective of the government (Republic of Kenya 1989b). About 96 percent of the mothers in Kwale district attended modern facilities for prenatal care. However, the majority of them (86 percent) gave birth at home, usually with the assistance of a traditional birth attendant. Health insurance is an integral part of a viable national system of user charges because it provides people with a mechanism for pooling resources against the risk of inability to afford the cost of medical care as well as against the risk of a financial ruin due to excessive medical expenses. Apart from the information shown in Table 2 (same sample sizes as in Table 5), an
64
GERMANO MWABU & JOSEPH WANG’OMBE
Table 5. Maximum likelihood estimation results [Dependent variable: Willingness to contribute to an insurance scheme (WILCOTRB_IN)] Explanatory variables CONSTANT QUALIT_PE INCOME_MO MALARIA_PR DISTANCE_KM ADULTS_HLD Sample size Log-likelihood a
Estimated coefficientsa Kirinyaga Kwale 0.589 (2.13) 0.224 (1.76) 0.001 (2.81) 0.124 (0.98) 0.003 (0.32) 0.044 (0.77) 1332 570.7
0.216 (0.82) 0.008 (0.71) 0.001 (0.85) 0.250 (1.58) 0.007 (1.17) 0.442 (1.01) 1338 923.7
Both areas 0.664 (3.29) 0.105 (1.29) 0.001 (2.23) 0.233 (2.64) 0.007 (1.61) 0.604 (1.78) 2670 1752.5
Absolute t-values in parentheses.
Statistically significant at the 10 percent level. Statistically significant at 5 percent level.
attempt was made to determine whether households in the study areas would be willing to participate in community based insurance schemes sponsored by the government. About 49 percent of the households in Kwale and 67 percent in Kirinyaga said they would be willing to contribute cash or in-kind in support of such schemes. Further, 95 percent of households in Kwale and 72 percent in Kirinyaga felt that insurance schemes should be arranged with government hospitals. We now consider the ordinary least squares regression results in Tables 3–5. Table 3 shows that income is the key determinant of private medical expenditure. The coefficient on income is positive and statistically significant for each district and for the two districts combined. A Shilling increase in household income raises medical expenditure by about 10 cents in the pooled regression, but notice that the slope of the expenditure function for Kirinyaga district is smaller than that for Kwale, being about 8 cents. The overall expenditure effect of malaria prevalence (Table 3, column 3) is positive and significant, indicating that households that experience frequent malarial episodes spend more on medical care. Households that often suffer from malarial attacks in the two districts combined spend about 6.8 cents
FINANCING RURAL HEALTH SERVICES IN KENYA
65
more on medical care relative to other households. In Kwale however, where malaria is a very severe problem, households that suffer from frequent malarial illnesses spend 4 shillings more on medical care relative to other households. In Kirinyaga however, frequent sufferers from malaria spend a mere 23 cents more on medical care relative to households only occasionally afflicted by this ailment. Another noteworthy finding from Table 3 is that medical care expenditure falls as the perceived quality of medical care rises. Turning to maximum likelihood results in Table 4, we see that household income is an important determinant of the willingness to pay for better quality care in government health facilities. The coefficient on income, although small, is statistically significant. Other determinants of the willingness to pay for service improvements in government health facilities include malaria prevalence and the number of adults in the household. In Table 5, income and malaria prevalence are the key determinants of the willingness to support a medical insurance scheme. Willingness to contribute to an insurance scheme declines with income but rises as the malaria prevalence increases.
6. Discussion of results In this section, we comment on some of the key findings of the study. Table 2 shows people’s overwhelming reliance on government health facilities for medical care in the study areas. This is perhaps due to high time and monetary costs of reaching alternative sources of formal medical care. Only about 25– 29 percent of households in the two districts were within 4–10 kilometers of a modern health facility at the time of the survey. On average, households lived 7–10 kilometers from the nearest health facility in both areas; about 64–75 percent of patients sought medical care from the nearest facility. Even though patients normally visited the nearest health facility, typically a government dispensary, there were occasions when that facility was bypassed. The most commonly cited reason for facility bypass was lack of drugs. This reason was given by 67–84 percent of sample households in the two districts. The next often mentioned reason for the bypass of the nearest facility was the long waiting time at the facility. Even though government health facilities were heavily used relative to nongovernment ones, people nonetheless spent considerable time to travel to any facility and to receive medical care after arriving there, a situation that must have discouraged some people from seeking treatment. On average, people spent more time waiting for medication than travelling to health facilities: the mean travel time in Kwale was 97.1 minutes [with a standard deviation of 63.5], while the mean waiting time was 151.1 minutes [with a standard
66
GERMANO MWABU & JOSEPH WANG’OMBE
deviation of 109.0]. In Kirinyaga, the mean travel time in minutes was 54.6 [52.1] while the waiting time was 169.8 [160.6]. It has been noted in the results section that most mothers gave birth at home with support of relatives or with the help of a traditional birth attendant, despite having attended modern clinics for prenatal care. Several reasons were given for this practice. About 28 percent of the mothers said that custom required them to deliver at home or that the traditional birth attendants were able to give them all the assistance they needed. Another 26 percent said they were unable to get to the nearest facility for delivery and a further 11 percent did not consider delivery to be a problem requiring medical attention. Despite their popularity among mothers for maternity services, traditional birth attendants charged about KSh 5–500 (US$ 0.25–25) per birth attended. This however does not imply that a similar charge on maternity services at government clinics would not reduce attendance there because the services being provided by traditional birth attendants might be of a different kind. In particular, traditional birth attendants in Kwale are all females by custom, an element that significantly differentiates their services from those of government clinics, where some of the midwives are men. The finding concerning the effect of frequency of malaria episodes on medical expenditure (Table 3) suggests that household health care budgets in malaria prone areas are more binding than in other areas. Indeed, the coefficient on MALARIA_PR (the extra medical expenditure induced by malaria prevalence in the village) can be thought of as an indicator of severity of malarial illnesses in the study areas or in any other region. A comparison of the coefficients on MALARIA_PR for Kirinyaga and Kwale districts shows that malaria prevalence is a more serious situation in Kwale than in Kirinyaga district. Notice also that the expenditure effect of malaria prevalence in Kirinyaga district is statistically insignificant. This however does not mean that this coefficient has no policy significance, for its statistical insignificance helps identify a region (Kwale in this case), where households face severe budget constraints in the financing of malaria treatments; the information should be helpful in determining the level of fees in the two districts. A key objective of instituting user fees in government health facilities, especially in sub-Saharan Africa is generation of revenue that can be used to improve service quality, primarily by improving drug availability. Drug availability at a health facility has been shown to be an important element in patients’ perception of quality of medical services, a factor that affects the rate at which these services are used (Wouters 1995). To assess without prompting households the importance of drug availability in the personal rating of service quality at government clinics, respondents were asked to state their worst experiences at the government clinics they had attended. About
FINANCING RURAL HEALTH SERVICES IN KENYA
67
52–62 percent of households in both districts mentioned lack of drugs as the most unpleasant experience they had at government health facilities. This suggests that an increase in drug supplies at these facilities would alleviate the problem of clinic bypass noted earlier. As to insurance schemes, people expressed the wish to use insurance mainly to meet the cost of hospital care probably because hospital services were perceived to be of a higher quality relative to services offered by alternative sources of treatment. Indeed, a missionary hospital, Chogoria Hospital in Eastern Kenya (not in study areas), has been quite successful in providing acceptable medical care to coffee farmers on a prepayment basis. People’s desire to tie medical insurance arrangements in rural areas with medical care in hospitals is a factor worth considering in the implementation of insurance schemes in such settings. It was observed from Table 3 that medical expenditure rises with income, a result that implies that the poor spend less on medical care than the rich. An obvious but important implication of this finding is that in a fee-for-service system, health care consumption in the population would be inequitable, a matter that should always be borne in mind when designing market-based mechanisms of financing health services. It has been noted in Table 3 that health care expenditure falls as the perceived quality of health care rises. One explanation for this finding is that better medical care reduces malarial sickness in the population thereby decreasing the number of visits to health facilities. Another possibility is that the quality of health care is positively correlated with the user charges for medical care, which are in turn negatively correlated with the number of visits to health facilities (see, e.g., Bloom et al. 1995). In other words, the quality effect on medical expenditure, which ordinarily should be positive, is mixed with an overwhelming negative effect of user charges. Because of data limitations no attempt was made to disentangle these two effects. The very low R-squareds in Table 3, indicate that the estimated expenditure relationship does not fit the data well. Nonetheless, the results are useful because the determinants of medical expenditures that they identify can be modified by policy to influence consumption of medical care. The index for the willingness to pay for government services in Table 4 (an indicator of the benefits expected from utilization of improved government services) declines as household income rises. This result implies that the probability that the poor are willing to pay for service improvement at government health facilities is greater than that for the nonpoor. This puzzling finding can be rationalized in several ways. First, the health status of the poor is in all likelihood lower than that for the rich so that the marginal health benefit due to better medical care is likely to be higher among the poor. Second, the poor might be relying on ill-equipped government clinics for medical care,
68
GERMANO MWABU & JOSEPH WANG’OMBE
whereas the rich are able to afford to bypass such clinics for better services in distant facilities, a situation that should make the poor relatively more willing to pay for service improvements in public clinics. Third, knowing that their resources are meager, the poor might be willing to accept the institution of fees because they expect the rich to bear a greater burden of fees without them controlling the use of improved services because the services are under public management and could thus be obtained free of charge. As a result of this potential unequal share of the cost burden of medical care between the rich and the poor, the expressed willigness to pay for better services at government clinics might not reflect the true benefits each of the two groups expects from such services. Thus, the effect of income on willingness to pay for improvements in government health services should be interpreted with caution. In this study, people’s willingness to support establishment of a medical insurance scheme rises as malaria prevalence increases (Table 5) because medical expenditures, which insurance helps to pay, are positively correlated with malarial episodes (see Table 3). This finding suggests that people are averse to risks of high medical expenditures in a malaria prevalent area. 7. Policy implications This study provides information of interest to health planners in low-income countries. An examination of pattern of visits to health facilities shows a strong tendency for people to use closest sources of medical care. Patients bypassed the nearest facilities for distant ones when they perceived some deficiencies in nearby facilities. Shortage of drugs was frequently cited as a deficiency of nearby government clinics. To reduce the potential for resource misallocation in service provision in rural areas (due to clinic bypass), priority should be placed on delivering accessible and acceptable services, rather than on quantitative expansion of rural health facilities. Household income emerges as a key determinant of people’s willingness to pay to improve government health services. The poor were more willing than the rich to pay fees at government clinics as well as to contribute in support of medical insurance schemes. This is a puzzling finding because the poor have a lower ability to pay. It implies that the general public would not protest against fees in a low-income area (as long as the reported willingness to pay is a true reflection of ability to afford fees). If however ability to pay is overstated, imposition of fees could engender a major public complaint, as was the case in Kenya during the initial phase of the health care financing reform (see Collins et al. 1996). Further, the ex-ante unwillingness to pay fees at public clinics expressed by the rich might be a strategy to avoid paying for a service that they value but which could be provided free of charge to some if it is considered
FINANCING RURAL HEALTH SERVICES IN KENYA
69
by policy-makers as a social good. If that is the case, both the rich as well as the poor would use public clinics after the fee introduction, with congestion at the clinics being a potential problem because service demand would likely have been underestimated. However, if the expressed unwillingness to pay by the rich reflects better alternative sources of care, the rich would not use public clinics after the fee introduction and the congestion problem would not arise. There is need therefore to use willingness-to-pay results from this study as well other studies with care. A major concern of financing health services through user fees in lowincome areas is that the fees might prevent a sizeable fraction of the population from seeking health care. This study reinforces this concern by showing that medical care expenditures rise with income, so that in a situation of fee-forservice, it is often the poor who would not afford medical care. This finding implies a need to exempt the poor from user charges in order to ensure equity in health care. It also implies a need to establish a system for collecting household data for use in updating the exemption criteria as the poverty profile of the population changes over time. The paper suggests a possible link between health care financing methods and the age structure of the population, arguing that user charges on curative services can alter the long run age structure of the population via spillover effects on preventive services. The limited evidence that we have reviewed shows that user fees for curative services may raise or reduce demand for preventive health care, such as prenatal and immunization services, depending on the nature of spillover effects of fees. To the extent that the use of such services affects fertility and mortality rates, the indirect and probably substantial effect of fees on age composition of the population cannot be ruled out. Acknowledgments We are very grateful to the Editor, Professor David F. Sly and two anonymous referees for helpful comments. The work reported here received financial support from Carnegie Corporation of New York and The Pew Charitable Trusts and also benefited from research facilities at the Economic Growth Center, Yale University. However, any errors in the paper are our responsibility alone. References Akin, S.J., Guilkey, D.K., Griffin, C. & Popkin, B.M. (1985). The demand for primary health services in the Third World. Totowa, NJ: Rowman & Allanheld.
70
GERMANO MWABU & JOSEPH WANG’OMBE
Akin, S.J., Guilkey, D.K., Griffin, C. & Popkin, B.M. (1986). Demand for primary health care in the Bicol region of the Philippines, Economic Development and Cultural Change 3(4): 755–82. Barlow, R. & Diop, F. (1995). Increasing the utilization of cost effective health services through changes in demand, Health Policy and Planning 10(3): 284–295. Bloom, E., Bitran, R., Dow, W., Straffon, B. & Orozco, M. (1995). The demand for health care in Mexico: An econometric analysis. Mexico: National Population Council, mimeo. Collins, D.H., Quick, J., Musau, S.N. & Kraushaar, D.L. (1996). Health financing reform in Kenya: The fall and rise of cost sharing, 1989–94. Management Sciences for Health, Stubbs Monograph Series Number 1, Boston, MA. Deaton, A.S. & Muellbauer, J. (1980). Economics and consumer behavior. Cambridge: Cambridge University Press. Dow, W.H. (1995). Welfare impacts of health care user fees: A health-valuation approach to analysis with imperfect markets. Economics Department, Yale University, New Haven, mimeo. Gertler, P. & Van der Gaag (1990). The willingness to pay for medical care: Evidence from two developing countries. Baltimore, MD: The Johns Hopkins University Press. Greene, W.H. (1991). Limited Dependent Variable Program (LIMDEP). Bellport, NY: Econometric Software Inc. Greene, W.H. (1993). Econometric analysis. Englewood Cliffs, NJ: Prentice Hall. Navaneetham, K. (1993). The influence of cohort effects on mortality trends in India: The role of economic factors, Population Research and Policy Review, 12(2): 159–176. Republic of Kenya (1989a). Introduction of cost-sharing in health service. Nairobi (Kenya): Official Circular No. B/13/118B, Ministry of Health. Republic of Kenya (1989b) Development Plan, 1988–1993. Nairobi (Kenya): Government Printer. Shaw, P.R. & Griffin, C. (1995). Financing health care in sub-Saharan Africa through user fees and insurance. Washington, DC: The World Bank. World Bank (1987). Financing health services in developing countries: An agenda for reform. Washington DC: World Bank. World Bank (1993). World development report: Investing in health. New York: Oxford University Press. World Bank (1994). Better health in Africa: Experience and lessons learned. Washington, DC: World Bank. Wouters, A. (1995). Improving quality through cost recovery in Niger, Health Policy and Planning 10(3): 257–270. Yazbeck, A. & Leighton, C. (1995). Does cost recovery for curative care affect preventive care utilization, Health Policy and Planning 10(3): 296–300.
Address for correspondence: Germano Mwabu, UNU/WIDER, Katajanokanlaituri 6B, FIN00160, Helsinki, Finland Phone: 358-9-615-9911; Fax: 358-9-615-99333; E-mail:
[email protected]