wPS I371/ POLICY RESEARCH WORKING PAPER 1874 Health Policy in Poor There is an app.-iViAnt consensus that tr e corre. Countries health policy in) teveioin countries is pub.ic provi -- * * . ~~~~~~~~~~~~~~arrmix of preventive anc! Weak Links in the Chain siple curative services through low lev&e heaith Deon Filmer workers and faculities. B,, the Jeffrey Hammer strength of th7i nsSen: , Lant Pritchett the 'primary hea;tn care paradigm' is in sharp c tr, n to either the st' rgthk oft analytical foundations ov-!fi mixed record ir pract,ce. The World Bank Development Research Group Poverty and Human Resources January 1998 S POLICY RESEARCH WORKING PAPER 1874 Summary findirngs Filrner, Hammer, and Pritchett show how the recent severity of market failures. Evidence suggests these are errmpirical and theoretical literature on health policy the least severe for relatively inexpensive curative sheds light on the disappointing experience with the services, which often absorb the bulk of primary health implementation of primary health care. They emphasize care budgets. Government policy in health can more the evidence on two weak links between government usefully focus directly on rmitigating market failures in spending on health and improvements in health status. traditional public health activities and, in more First, the capability of developing country governments developed settings, failures in the markets for risk to Provide effective services varies widely - so health mitigation. Addressing poverty requires consideration of spending, even on the "right" services, may lead to little a much broader set of policies which may - or may not actual provision of services. Second, the net impact of - include provision of health services. government provision of health services depends on the This paper-a product of Poverty and Human Resources, Developrnent Research Group - is part of a larger effort in the gro-up to investigate efficacy in the social sectors. The study was funded by the Bank's Research Support Budget under the research project "Primary Health Care: A Criticai Examination" (RPO 680-29). Copies of this paper are available free from the World Bank, 1818 H StreetNW, Washington, DC 20433. Please contact Sheila Fallon, room MC3-638, telephone 202- 473-8009, fax 202-522-1153, Internet address sfallon@wortldbank.org. January 1998. (63 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of Ideas about development isi-s. An objective of the series is to get the findings ozlt quickly, euen if tbe presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the vietw of the Wlorld Bank, its Execsttiue Directors, or the countries they represent. Produced by the Policy Research Dissemination Center Health Policy in Poor Countries: Weak Links in the Chain Deon Filmer Jeffrey Hammer Lant Pritchett The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. This paper grew out of an earlier collaboration between the authors and Maureen Lewis and Samuel Lieberman. We would like to thank Martin Ravaillon and Susan Stout for helpful discussions. Comments are welcome. I Health Policy in Poor Countries: Weak Links in the Chain Introduction Arguments for "primary health care" (PHC) are very appealing as the problems it addresses are pressing and the solution it provides seems obvious. In 1995 over 9 million children under five in developing countries died avoidable deaths; more than the entire population of Sweden or of Zambia.' There are developing countries whose budget is principally absorbed by public hospitals staffed by extensively trained (at public expense) doctors using expensive medical technologies to treat conditions of the urban elite, while in those same countries children die from diseases that could have been treated for a few cents or avoided altogether with basic hygienic practices. Yet there are examples of success. Kerala, a state in India with income per capita of only $12542 has infant mortality of only 31 (per thousand live births), which is not only forty percent lower than that in Punjab, another Indian state with twice the income, but is 35 percent lower than that in Brazil, with over four times the per capita income. Infant mortality in Shanghai is lower than that of Manhattan and the recorded infant mortality rate in Jamaica of 16 is lower than that of African-Americans in the United States. Ceara, one of the poorest states in Brazil reduced infant mortality by 36 '"Avoidable" deaths are defined as the excess of the average death rate for the 0-5 age group in the low- and middle-income countries of 88 per 1000 versus the level in the high- income countries, 9. Using a comparable approach Gwatkin (1980) calculated the total number of deaths of under fives to be about 15 million. 2 Income is in 1995 intemational dollars which are adjusted for purchasing power differences. 2 percent in just a few years through an aggressive government program (Tendler and Freedheim, 1994). This combination of experiences has led to a strong two-fold consensus on health policy in the poorest countries.3 First, economic progress is not enough and governments can, and should, act to improve health. Second, the existing allocation of health expenditures towards curative care in tertiary facilities is inappropriate and a reorientation of government efforts towards PHC would bring both health gains and cost savings. In this consensus PHC is typically defined expansively. PHC could be defined by what it is not: it is neither secondary nor tertiary curative care, but all other activities related to health, from nutrition to sanitation. Even more ambitious definitions view PHC as a part of social revolution (Decosas 1990). For our purposes we will treat PHC as composed of three (at least conceptually) distinct elements: simple curative care usually based in "primary" facilities, preventive activities aimed at health imnprovement especially those based on community health workers, and finally more traditional public health campaigns.4'5 I The current debates on health care reform in the developed countries, and in the upper middle income countries with similar health profiles, such as Eastern Europe and the richer parts of South America are rightly focused on completely different issues. In those countries, the epidemiological profile of mortality and morbidity more towards non-infectious diseases of adulthood, such as heart disease and cancers, which have high, and highly skewed treatment costs per episode. This lends to a greater focus aggregate cost containment, relationships between medical innovations and costs, and relationships between patient, provider and payment. 4 These are similar to the "promotive, preventive, curative, and rehabilitative services" of the Alma Ata declaration (WHO, 1988). 5 Many see a "basic package of cost-effective interventions" advocated by some at the World Bank (Bobadilla and Saxenian, 1993) as a minor variant on PHC. This is not exactly true, as the technique of "medical intervention cost effectiveness" (MICE) has two differences with 3 While the images and statistics that motivate PHC are compelling, the gains have rarely been demonstrated in practice. Although there have been huge successes from individual campaigns under a PHC banner including various immunization campaigns, or the recent campaign against river blindness, the data often show very little impact of PHC. The implementation of PHC has created a new set of images. Empty rural health clinics without drugs or working equipment. The sick bypassing free primary public clinics to pay for services from private providers. This disappointing experience of raises the question: what was missing from the seemingly compelling logic? We believe economists bring two perspectives that are useful for understanding, and perhaps improving, the implementation of health policy: choices and incentives. First, too often the impact of PHC was calculated as if health status were entirely a technocratic affair and individuals were the passive recipients of government action. But individuals actively use their knowledge and resources to enhance their own (and their children's) health. Incorporating choices into the analysis can completely change both the expected impact of PHC and the ranking of the importance of various actions. The impact of PHC cannot be assessed from medical knowledge, but depends both on how it impacts on the demand for services and on how it interacts with the existing (and potential) supply and prices in the private sector. PHC. First, at least in theory the level of facility does not matter as high MICE interventions could occur in secondary or tertiary facilities. Second, MICE analysis is also claimed to be useful as a technique for cost containment in higher income contexts where PHC is less relevant. The problems with MICE as a tool for public policy are serious but addressed elsewhere (Hammer and Pritchett, 1997). 4 Second, PHC advocates often assumed the public sector could be made to deliver whatever was decided in the capital (or at an intermational forum in someone else's capital) ought to be delivered. In practice, the quality of public sector health services has ranged from excellent to truly horrific. While an idealized, well run network of community workers and rural health clinics might have a dramatic impact on health status, the real issue is the impact of the services a country's public sector is actually capable of providing. Public sector failures in health are not just random but are results of a systemic mismatch between the incentive structure in the traditional civil service mode of public sector organization and tasks in the health sector. The paper is organized as follows. Section 1 presents a simple framework to organize the literature. Sections 2 reviews the cross-country evidence on the small impact of public spending and discusses whether this is only because public monies have been spent on the wrong type of services. The subsequent sections take up possible explanations of the impact of PHC. Section 3 discusses the organizational structure of the public sector in health and how it might lead to public spending creating ineffective services. Section 4 discusses how public and private interactions might mitigate the impact of public intervention. Section 5 looks at the role PHC might have in the presence of market failures and in the context of the links between poverty and health. I) A simple framework for analyzing the lIterature We start with a very simple story of individual maximizing behavior, one which motivates a chain of reasoning about the potential impact: of public spending on health. 5 Suppose an individual has an exogenously given total resources (1) and level of knowledge (E), and his or her well being is determined by health status (HS) and the consumption of all other goods (X), whose price we normalize to 1. Suppose that health status is produced from two health inputs, HI and H2 whose prices are p, and P2 (the prices include not just money cost but all resources sacrificed) and whose effect is conditioned by the individual's knowledge (E). The individual's choice problem is to choose the levels of consumption to maximize their welfare subject to the budget constraint and the health production function: Maximize U(X,HS) with respect to X, HI, H2 subject to: X + P1xH1 + p2XH2 = Y and HS = HS(H,,H2;E) The individuals' solution to this problem gives the optimal consumption levels, X*, HI* and H2*. Substituting those into the health production function gives the individuals' health status as a function of their income, knowledge, and the relative prices of the health goods, HS*(Y, E, pl, p) (where "*" represents the optimal quantity). There are three questions that the literature has addressed: one positive, one normative, one practical. The first is, how big is the effect of any given type of public spending on health (d HS*/d PSI) ? The normative question is, if a benevolent dictator were trying to maximize 6 health status using the allocation of a fixed budget, how should she allocate public spending?6 The third, and most important, is the practical question: what should be done? Our simple framework helps us organize these questions. Public spending influences health status by lowering the effective price of health enhancing inputs. How it does so depends on four distinct mechanisms, which can be expressed as a chain of partial impacts that lead from public expenditures to improvement in health: d HS - aHs aH1 a Qs a PSI x - x . x d PS 8 H' QPS aPs1 a PS + aHS. a'H a PSaPs2 x 2x 92xa S a H1 aQ_ aPS2 a PS + indirect and cross -effect terms 1) cl PS, / a PS. Composition of public spending. The impact of an increase in total public spending on health will depend on how that increase is allocated across health inputs. An equi-proportionate increase in spending on all inputs will have a very different impact than one which increased only the most effective public interventions. 2) a Q/' / a PS, Public sector efficacy. Even when money is spent, the question is whether it creates effective health services, and in what magnitude. When a govermnent decides to build a clinic, or spray for malaria, or mobilize community outreach workers, or buy X-ray 6 Along with the nearly all of the rest of the literature, we assume a fixed budget and avoid the problem of maximizing welfare as that would endogenize the budget and require valuing health versus non-health goods. 7 machines, it can be more or less effective at translating that expenditure into a real supply of services. This efficacy will have both a country specific component common to all activities, but also an activity specific component, as govermnent might be more or less well adapted to certain activities. 3) a H, / a QtPS: Net impact on use of services of public sector supply. Even if public sector expenditures do create a supply, the next question is how this expansion translates into a change in the effective price faced by consumers. Even if a particular health service is "cost effective" in improving health, this does not mean that public spending on that service would be cost effective in improving health, as additional consumption in the public sector occasioned by public supply may well simply crowd out, in whole or part, equally effective services obtained from non-government providers. The size of this effect will depend on the responsiveness of individuals' demand, and private suppliers' supply, to changes in the price, travel time, convenience, or quality of services, induced by changes in public availability. 4) a HS* / a Hi*: The Health Production Function. Different health inputs are more or less effective in improving health in ways determined by biological and medical facts. What the health production function looks like, that is, which treatments are effective in eliminating which cancers, which vaccines are potent over what period, how micro-nutrients affect susceptibility to diseases, is what health care professionals learn. Economists typically prefer to remain agnostic about the production function particulars and in many cases sensible recommendations about public policy need not necessarily inquire into the production 8 function.' In this framework, the argument is that increases in public spending on PHC are effective in improving health status, while curative and especially secondary and tertiary curative services are not. This can rationalize an increase in public funds spent on PHC as well as a reallocation of the health budget towards PHC activities. The argument for PHC, however, typically relies almost exclusively on the health production function (perhaps, modified to include accounting costs to generate MICE rankings). However, the actual impact of public spending is the product of all four terms in the above equations: allocation of the budget, public sector efficacy, market impact on consumer demandfor services, and health impact of services. If any one of these is low, the total impact will be low. II) Impact of public spending and PHC on health status Has public spending on health, and more particularly on PHC, promoted good health? We review three strands of evidence: first, country level evidence, second, the impact of facility availability on the health status of individuals, and third, the evaluation of projects and experiments. Mortality is easy to measure while morbidity is not and (fortunately for people but unfortunately for research) mortality (except among the very young and very old) is rare. 7 One prominent economist when asked what an economist needed to know about the particulars of the production function in order to make sensible policy recommendations, responded "Convexity." 9 Wbile WHO's definition of health as "a state of complete physical, mental, and social well- being, and not merely the absence of disease or infirmity (WHO, 1988)" is attractive, it is subjective and hard to assess.8 Moreover, since death occurs rarely, and only once for each person, it is difficult to study at the household level. So while we might be interested in the totality of health over an individual's life course, empirical studies tend to focus on infant (or child) mortality or life expectancy as proxies for health status and rely on aggregate (district, province and country) comparisons. Country level: Aggregate spending. Cross national studies have come to a fair consensus on two points. First, socio-economic characteristics explain nearly all of the variation in mortality rates across countries. A recent cross-national econometric study of child (under-5) mortality shows that average GDP per capita, a measure of the distribution of income, the level of female education, a dummy variable for countries predominantly Muslim, an index of ethnolinguistic diversity, and a set of five dummy variables for regions explain virtually all the variation in child mortality (Filmer and Pritchett, 1997). GDP per capita alone "explains" 80 percent of the variation in mortality and adding the other variables raises this to 95 percent. Preston (1980), in an influential paper based on data from between 1940 and the 1970 emphasized the low explanatory power of socioeconomic variables. However, more recent data and results are unanimous about the high explanatory power of socio- economic basics like average income and female education, including Preston (1986) using 'The huge effort to create measures of Disability Adjusted Life Years (DALYs) has led to some additional information on morbidity (Murray, 1994). However, the correlation between DALYs lost and life expectancy or infant mortality is 0.93 across the eight regions for which DALYS have been calculated (World Bank, 1993). 10 data from between 1970 and 1980.9 Second, total public spending on health has had much less impact on average health status that one might have expected, and certainly less than hoped. Although the lack of data on public spending has, until recently, limited the direct examination of the issue, Musgrove (1996) summarizes studies of the impact of public spending on health on health status: "[m]ultivariate estimates of the determinants of child mortality give much the same answer [as his results on life expectancy or DALY burden of disease]: income is always significant, but the health share in GDP, the public share in health spending, and the share of public spending on health in GDP never are." Even using instrumental variables to account for data and endogeneity problems Filmer and Pritchett (1997) find that public expenditure on health as a share of GDP is a small, and statistically insignificant, determinant of child mortality. At the point estimates doubling public spending from 3 to 6 percent of GDP would improve mortality by only between 9 to 13 percent (appendix Table A-I reproduces those regressions). 1 Other empirical studies that add measures of health resources, such as physicians, nurses, or hospital beds per capita to the basic socio-economic variables in this type of regression rarely find large and significant impacts for these variables (Kim and Moody, 1992). 9 Some have been confused about Anand and Ravallion (1993), claiming that they showed that income was unimportant, in spite of their explicit claim otherwise. What they showed is that they could not reject the restriction that average income affected health status only insofar as it affected the level of poverty, highlighting the fact that it is increased income among the poor that is most effective in improving health. However, since the correlation between average incomes and poverty is very high this still implies average income, through its poverty reducing effects, explains most health status variation. 10 Bidani and Ravallion (1997) show a large impact of public spending on the health status of the poor, but their estimated impact of public spending on aggregate health status (of the poor and non-poor taken together) was also quite small. 11 So far the evidence is consistent with the following argument for PHC. First, although socio-economic conditions powerfully determine health status, there are still outliers such as Kerala, Sri Lanka, Costa Rica whose achievements are potentially replicable. Second, there are simple health "interventions" that could be delivered in basic facilities that would avert a large fraction of the deaths in low income settings at very low cost, typically between $10 and $4000 per death averted (Jamison and others, 1993). '1 Third, the evidence of the low effectiveness of existing public spending, such as the finding that for the typical country $50,000 to $100,000 is spent per death averted (Filmer and Pritchett, 1997) is "proof" that reallocations of the existing public budget to PHC could lead to large health gains at no cost (or even with savings). However, while this argument appears to create a powerful presumnption that a reallocation of the public budget towards PHC would significantly improve health status, it does not demonstrate this. What is the evidence? While the existence of health "outliers' suggests possibilities, it was never very clear the success of the outliers was due to a health system versus social or political phenomena. A participant at a seminal conference which cemented support for PHC based on case studies of " The MICE literature is usefully broken into two types: bottom-up or top-down, based on the epidemiological conditions envisaged. MICE concerned with health expenditures in very low income countries is a "bottom-up" approach to building a "basic package" and tends to reinforce PHC recommendations (with some exceptions, as some hospital based clinical treatments are high MICE). MICE in high-income countries, which face an entirely different pattern of disease conditions is "top-down": geared to limiting costs by eliminating payment for extraordinarily low MICE interventions (e.g. heart transplants in the elderly) and is of limited relevance in most low income settings. 12 the "outliers" commented:'2 The four case studies [China, Costa Rica, Kerala state, Sri Lanka], involve societies in which low mortality has been reached without high per capita income. Situations in which low income continues to be associated with high mortality or high income is associated with high mortality were not considered, nor have we searched systematically for other societies in which relevant social characteristics of the four successful cases are repeated, to see what happened to mortality. Thus the policy prescriptions are relatively weak. (Kunstadter, 1985, p. 234) Moreover, while the "barefoot doctors" of China are famous, it is not obvious all successes followed a "PHC" like strategy. In 1986 Sri Lanka spent 70 percent of its public monies on hospitals, substantially higher than the 56 percent average for comparable countries in South Asia (Griffin, 1992). If the reasoning behind the recommendation of PHC-like activities is correct then there should be empirical regularities both at the aggregate and local levels. First, given the total level of expenditures, more spending on "PHC-like" activities and greater access to "PHC- like" services should be associated with lower mortality. Second, at the locat level (household, village) we should see a positive association between greater access to health care facilities of the PHC type and lower mortality. Third, projects which create facilities should lower mortality. None of these regularities find much support in the data. Country level: PHC. Table 1 reports the results of including either the share of national health expenditures that are devoted to local health services or access to local services (defined as the share of the population with local health services, including essential drug availability, 12 This was a conference sponsored by the Rockefeller Foundation whose results are documnented in Halstead et al (1985). 13 within one hour's walk or travel) in an equation explaining child mortality."3 The results in column 3 and 4 show that mortality is not significantly systematically lower where more spending is directed at local health services. The results in columns 5 and 6 show that mortality is not significantly systematically lower where populations have greater access to local health services. While it is easy (and many times correct) to dismiss cross-national regressions as definitive evidence against any particular claim, as the "true' variable of interest might be badly measured, it must be said that the cross national evidence has not yet been marshaled which lends support to PHC.14 '3 This is into a cross national regression which explains under-5 mortality with income, public spending on health as a share of GDP, female education, income distribution, and other non-health related factors. A description of the data and the full set of results are in Appendix A-1. 14 Two stage least squares estimation was used to address the potential problems of measurement error and reverse causation. The estimates will be biased towards zero if these variables are measured with error (this is likely as a value for 1985 is sometimes used in place of one for 1990). In addition, the esmats will be inconsistent if there is reverse causation, for example if high mortality cause a government to spend more on providing access to more local health services. See appendix Table A-1 for the instruments used. 14 Table 1: Selected coefficients from under-5 mortality rate (In) regressions (1990) Column: 1 2 3 4 5 6 Method OLS 2SLS OLS 2SLS OLS 2SLS GDP per capita (In) -.611 -.596- -.624 -.921- -.616 -.829- (9.71) (3.67) (8.16) (1.84) (9.08) (2.47) Public health exp. -.135 -.192- -.165 -.139- -.085 .146- (Ln share of GDP) (1.78) (.742) (1.73) (.359) (.931) (.374) Local health .076 -.359- services (In share (1.78) (.754) of expend) Access to local -.024- .336- health services (In (.227) (.417) share of pop) Ad Ftonal Female education, income inequality, percent urban, dummy for predominantly Muslim, vadiiblea. ethnolinguistic fractionalization index, dummy for "tropical" country, access to safe variables. j water, dummy variables for region and a constant term. R-squared .9469 F 9465 .9512 .8834 .9608 .9430 Num. Obs. 98 98 73| 73 75 75 Notes: White heteroskedasticity-corrected t-statistics are in parentheses. -Instruments are, neighbors' public health spending, neighbors' military spending, whether or not the country's main export is oil, and years since 1776 that the country has been independent), and neighbor's shares of spending on local health (column 4), and access to local health services (column 6). See Appendix A for full results. Local outcomes. A second empirical regularity that would be supportive of PHC is if the availability of primary level health facilities or community health workers had a demonstrable impact on local health status. However, the results on the effect of access to hospitals, doctors, and in particular here, public sector clinics, health centers, and rural health workers on health status is, at best, mixed. A huge technical problem with empirical assessment of the impact of health facilities is that governments may have systematically placed health facilities. If the government places clinics where health status is worst, then a comparison of health status in localities with and without clinics would understate the true impact of clinics. Conversely, if the government 15 places clinics in villages where the population articulates the greatest demand, these may be where health would have been good in any case, and therefore comparisons of health status in localities with and without a facility would overstate the impact. Some recent studies assess the effect of access to services on child or infant mortality using methods that take into account that the placement of facilities or services may vary systematically in response to local characteristics. Frankenberg (1993) controls for placement effects using a sample of pairs of randomly matched births from two different cohorts in a village in Indonesia and finds the presence of a maternity clinic or of a doctor reduces mortality, but also the presence of a health worker increases the probability of death (statistically significant only at the ten percent level). Pitt, Rosenzweig, and Gibbons (1993) address the placement effect using a panel of matched districts in Indonesia and find that the share of villages in a district with a health center increases mortality (statistically insignificant) while the share of villages with a family planning clinic reduces mortality (also statistically insignificant). There are many econometric studies of the impact of faculties which do not control for selective placement.'5 Panis and Lillard (1994), after controlling for the potential endogeneity of facility usage, find that delivering a baby within an institution, the likelihood of which increases with facility availability, reduces the probability that a child would subsequently die in Malaysia but find, puzzlingly, that the use of prenatal care insignificantly increases the IS In her uncorrected estimates Frankenberg (1993) finds that more maternity clinics and health workers insignificantly reduce mortality while more doctors insignificantly increases mortality. In their uncorrected estimates, Pitt et al (1993) find that the presence of both health centers and family planning clinics raise mortality (insignificantly in the case of health centers). 16 probability that the child will subsequently die. Benefo and Schultz (1994) find that households further from a clinic have higher child mortality in Cote d'Ivoire, but in Ghana proximity to clinics appears to increase mortality (statistically insignificant). Lavy, Strauss, Thomas, and de Vreyer (1996) find that the distance to health facilities significantly decreases mortality in rural Ghana (the presence of child services in that clinic is also significantly positively related to survival, however). In Malaysia, DaVanzo (1984) finds distance to medical care does not reduce infant mortality conditional on birth weight, but that birth weight is lower the greater the distance to care. Hossein (1989) finds that the presence of a dispensary and presence of a family planning clinic lowers mortality in Bangladesh. Rosenzweig and Schultz (1982) find that in rural Colombia, neither rural health posts, municipal level public and private clinics, dispensaries, or mobile care units are significantly related to child mortality. For urban areas they find that hospitals, clinics, and family planning clinics tend to reduce mortality, however this result is not consistent across all age groups. Rosenzweig and Wolpin (1982) use data from rural India to find that the fraction of villages in a district with a family planning clinic and the fraction with a dispensary, is associated with lower mortality, but that the fraction with an "other health facility (health centers, nursing homes, etc...)" is associated with higher mortality. Sastry (1995) finds that the number of general health facilities are insignificantly associated with higher mortality in Northeast Brazil and insignificantly with lower mortality in South/Southeast Brazil. Overall, the econometric evidence is mixed that there is a health effect of clinics at all, no less that it is large. Micro level: Project Evaluation. There are a few evaluations of systematic 17 experiments. Beginning in 1977 a very intensive program of provision of Maternal and Child Health and Family Planning (MCH-FP) services was introduced in a set of treatment villages in the Matlab region of Bangladesh (with a nearby set of villages served as a comparators). Mothers and children were visited every 15 days in the treatment area by a female worker with messages about family planning. Detailed vital records were kept in both areas. Although mortality among children fell (Muhuri and Preston, 1991), this has largely been attributed to measles immunization (Koenig, Faveau, and Wojtyniak, 1991, Menken and Phillips, 1990). An experimental design of the delivery of health care services for children which was intensively carried out in Narangwal, India, showed a six point drop in infant mortality over the three years of the project (from 96 per 1000) versus a one point increase in the control area (from 107 per 1000) with the difference being insignificantly different from zero (Taylor and Singh, ND, page IV.D.7).'6 Taken together, these empirical results suggest that enhancing health outcomes is not simply a matter of providing additional funds, or increasing access to PHC-like services and facilities. This lack of demonstrated impact of PHC across a variety of countries and settings raises the importance of the framework discussed in Section 1, as the "chain" provides two likely explanations for why the impact on health status of public spending, even on PIM4ike interventions, might be low. First, the impact of provision will depend on the ability of the 16 latter rate is, by the authors' admission, a poorly measured one as it is based on a post-neonatal mortality rate increase in the three years of 49 percent. The authors explanation of this huge fluctuation is "due to incorrect age classification, [...] since in the control villages [...] vital statistics investigators who had no access to exact birth information and needed to rely on the age at death given by the mother of the family". 18 public sector to create effective services. Second, the impact of the provision of public services will depend on choices and the market for health, that is both on the private demand and how it may vary across disease conditions, and private supply and how it responds to public intervention. III) Public sector spending and the creation of effective health services One reason why PHC might have little impact on health status is not that in principle PHC is unimportant, but that in practice the efficacy of government actions has been low. Without personal experience it is difficult to appreciate how bad "low quality" public sector services can be. Traveler's "tales" of public health clinics are legendary and we'll give just three anecdotal examples. First, while one of the authors was visiting one low income country a prominent newspaper accused the Ministry of Health of misappropriating $50 million of donor financing. The ministry the next day accused the newspaper of exaggeration and irresponsibility for failing to make it clear that this $50 million was misappropriated over a period of three years, not in a single year as the newspaper reports implied. Second, a client survey of women who had a birth in the past 2 years at rural health centers in the Mutasa district of Tanzania revealed the most frequently cited disadvantages of giving birth in an institution were: ridiculed by nurses for not having baby clothes (22 percent), maternity fees (16 percent), nurses ordered mothers to wash linen used soon after delivery (16 percent), and nurses hit mothers during delivery! (13 percent) (Mtemeli, 1994). ' Third, in nearly every '' Interestingly, when nurses themselves were asked to give reasons why they thought mothers did not deliver in health institutions, the most frequently cited reasons were: 19 country one can find rural health clinics completely without drugs while the government (or donor) financed medicines are easily available on the black market. For example, over 70 percent of the government supply of drugs disappeared in Guinea in 1984 (Foster, 1990). Various studies in Cameroon, Uganda, and Tanzania estimated that about 30 percent of publicly supplied drugs were misappropriated, in one case as much as 30 to 40 percent of public supply was "withdrawn for private use" by staff (World Bank, 1994a). A) Evidence on quality There are a number of empirical studies of how quality of is linked to demand for public facilities (e.g. Akin, Guilkey, and Denton, 1995, Lavy and Germain, 1994, Lavy, et. al., 1996, Mwabu, Ainsworth, and Nyamete, 1993, Thomas, Lavy, and Strauss, 1996) and this literature is well reviewed in Alderman and Lavy (1996). Even though the measures of quality are not always satisfactory, the findings are that demand is responsive to quality. An example of the problematic nature of some of the measures is that the absence of various types of drugs is often used to indicate poor quality. Shortages, however, could be caused by high demand and hence it is hard to draw inferences about the causal relationship (a problem generally acknowledged by the authors). In addition, there may be important discrepancies between de facto and de jure measures of quality, a result highlighted by Thomas, Lavy, and Strauss (1996) who show the differential impact on health outcomes of the actual versus the official number of staff. distance/transport problems (20 percent), lack of clinic fees (14 percent), harassment by nursing staff /fear of nurses (1 1 percent). 20 Both the ineffectiveness of low-level public sector health clinics and the element of individual choice are highlighted in the phenomenon of "bypassing." People often do not go to the closest public facility but "bypass" it in favor of either more costly private facilities or higher level public facilities. There are few empirical studies of bypassing as this requires detailed information on both the health seeking behavior by individuals as well as information on all potential sources of supply. A recent study in Sri Lanka uses detailed surveys of health care supply and demand to document large amounts of blypassing (Sarasinghe and Akin, 1994). Table 2 shows that of all illness episodes only 29 percent are treated at the closest facility (which are predominantly Ayurvedic and "minor public western"), as 31 percent "self-treat" with no medical consultation and, of those that seek some treatment, 58 percent do not go to the nearest facility.'" Of the non-Ayurvedic sources, the "minor public western" facilities, the type that PHC would promote, are bypassed the most. Most individuals who bypass either "minor" or "major" public facilities do so to visit a "private western" facility. Table 2: Summary of bypassing results of Sri Lankan study 1 2 3 Treatment chosen by ill Percent of times each type of For each type of facility in column Self-care 31% facility is bypassed 2, the main type of facflItybat the Closest facility 29% bypassing individuals go to Ayurvedic 40% Private western 41% Not cls f40% Minor public western 21% Private western 50% Not closest facility Private westein 10% Major public western 58% Major public western 3 % Private westem 68% Source: Adapted from Sarasinghe and Akin, 1994 The authors fimd the public facilities bypassed have fewer doctors, nurses, services, '9 Ayurvedic are the local "traditional" sources of medical treatment. 21 and levels of equipment. In contrast, the private western facilities that are bypassed are those with more doctors, nurses, services, and levels of service. While this might seem paradoxical, this is consistent with sophisticated health seeking behavior on the part of individuals. Since prices tend to be lower in the public facilities they will bypass sophisticated, but expensive, private for public if the condition is not serious or quality is not important. That is, they will bypass "expensive relative to the disease condition' private western facilities. However, for serious conditions or when quality of service is important, individuals will be willing to pay-- both in terms of time and fees-for higher quality care. Bypassing leads to low utilization of available public facilities. A recent survey of a rural area of Punjab province, Pakistan, found that although the physical infrastructure of rural PHC was in place "[o]nly about 5 percent of the sick children were taken for treatment to primary health care facilities; half were taken to private dispensers, and another quarter to private MBBS doctors. Around 95 percent of deliveries took place at home." (PIEDR, 1994, P.vi). Roughly the same percentage of respondents sought treatment from a public rural health facility (5.2 percent) as a "quack" (4.9 percent) (PIEDR, 1994, p35). This was not because individuals were deterred from public facilities because of queues--on the contrary, the typical rural health center was seeing only about 30 patients a day, and the typical "basic health center" only 11 patients a day which was far below capacity as rural health centers had on average eight workers, and basic health units five. Two atypically busy rural health centers attracted and serviced an average of over 450 visits per day. Low utilization is attested in a recent study of health center use in Indonesia. Even 22 where public facilities were located in close proximity to large local populations annual caseloads were low (World Bank, 1994b). Based on detailed case studies the study identifies two principal reasons. First, many public facilities were lacking equipment, medicines, and appropriate health workers. Second, and more importantly, detailed assessments on the way resources were combined to produce health outcomes showed that poor functioning contributed to a large degree to the low public facility utilization. Respondents in one case-study felt that public facilities were of low quality while "[tihey were confident that they could get considerate and unrushed care in a pleasant and informal setting in the private practice of doctors, bidans [midwives] and nurses" (World Bank, 1994b). A recent study in El Salvador found remarkably similar results. Respondents consistently complained about the low quality of public health posts and units, especially relative to higher level services from health centers and hospitals. Typical answers were along the lines of: Health posts operate only twice a week. Consultation is only until noon. The doctor is not always there. Sometimes only the nurse assistant is present. Waiting time is three hours on average. Only those who arrive by 8 get a consultation (World Bank, 1997). In contrast to, for example, one typical respondent in El Pinar "[The health center at] La Palma is a little hospital with very good services. It is well equipped. The fee is only c/3 for consultation and sometimes medication" (World Bank, 1997) The study in El Salvador assessed the impact of the very lowest level intervention, the health promoter who lives within the community and is supposed to be the first level of information and referral. The assessment found that focus group respondents had very little use for these promoters, and regression analysis found that health promoters from the public sector had little 23 or no effect on the probability of seeking medical treatment. Even when the care provided in the public sector is of reasonable quality, it may be tremendously inefficient. An stark example of the kinds of inefficiency possible in the public sector is from an extremely detailed study of the expenditures versus the actual costs of production in a public hospital in the Dominican Republic measured through careful observations on time use (Lewis et al, 1996). The study showed that, although spending on personnel constituted 84 percent of total recurrent spending, actual staff costs were only 2.5 percent for treating emergency patients, 5.1 percent for inpatients, and 11.5 for outpatients (each expressed as a percentage of total costs).'9 Gross inefficiency was identified as the cause for this huge discrepancy. More importantly, the study concluded that the causes for these extraordinarily high costs lie within the functioning of the- hospital. There was no accountability for physician or nurse performance, no rewards for extraordinary performance, no punishment for inadequate or nonexistent performance, salaries were low and undifferentiated, there was no management control over staff, and essentially no returns to effective management20. A9 comparison of public and private hospitals in Argentina found that when accounting for all services the total was 2.4 services per professional per day in the public sector versus while in a private hospita doctors ~Mrried out 2.4 consultations per hour. 20 There are nearly always serious problems with the private sector as well. But the underutilization, pure waste of materials, and ordinary inefficiency discussed above rarely make for a living wage for a professional, which creates a powerful incentive for efficiency. 24 B) How is quality possible in the public sector? Even strong advocates of PIIC would agree that there have been egregious failures of governments to provide high quality health services cost effectively. Conversely, even the most ardent critic of government acknowledges that there are examples of admirable and well functioning health facilities and agencies in the public sector. The difficult issue for health policy is whether failures are the result of ignorance and mistakes or whether failures are a systematic and expected result of incentives created by institutional and organizational arrangements. If one believes that existing public sector problems are easily remediable: through larger budgets, or eannarked inputs, or additional training, or "technical assistance' of various kinds, then the drawbacks of the quality of the existing public sector is no reason to back away from public provision as a strategy. On the other hand, if one believes the failures are endemic and intrinsic to the public sector one needs to reassess the strategy for the delivery mechanism of health services in its entirety. Pay, employment, and performance in the health sector. There is no one right answer about public sector capability in the health sector. Somie public sector agencies provide high quality, cost effective, health care. Other public sector agencies are capable of spending unlimited amounts of resources with no health gains. The tough policy questions are (1) are the conditions in place for effective public provision of particular health services? and if not then (2) is it possible to achieve those conditions within a reasonable time horizon? While some critics have perhaps been overly pessimistic about government capacity, supporters of publicly provided clinical services have paid far too little attention to the first question and been entirely too sanguine about the second. 25 The feasibility of incentivizing pay and employment in the public sector is not a new issue. It is generally recognized that the more essential and the less easily observable individual effort, the greater the importance of linking pay and performance (Milgrom and Roberts, 1992). Workers who are in situations where effort is easily observed and monitored are usually paid wages or salaries, while those working where output is observed but effort is not, like salesmen, are paid on the basis of outcomes. But in addition to the level of pay there is the question of continued employment. Where output is crucially linked to individual performance and there can be little tolerance for deviations from high quality, then continued employment is generally linked to performance. Observation of pay and employment across the public and private sector tends to reinforce this position. As illustration, Table 3 shows a matrix of jobs and how closely pay and employment tend to be related to performance. Table 3: Various types of employment and compensation schemes Degree to which pay is linked to performance Low Medium High Degree to which Low Traditional civil Stable large Piece Rate employment is linked service organizations (harvesters, to performance arrangements (e.g. salesmen, postal workers, contract administrators) workers) High "Up or out" Most private sector Professionals organizations (e.g. organizations (e.g. law firms, US military) medical practices) A glaring feature of this table is that both the pay and employment of private sector professionals (for example, lawyers, or doctors in medical practices) tend to be highly related to performance while in traditional public sector organizations neither are. While such an 26 observation is perhaps a commonplace, it has rarely been questioned that the provision of clinical services by health care professionals is amenable to public sector organization, even though quality is essentially unobservable to outsiders (both in terms of health efficacy and of client treatment). One reason for this lack of inquiry is that sometimes the most obvious and seamlessly working features of a system are invisible and taken for granted when the system is functioning well. As a consequence, when one attempts to extrapolate from one set of social, legal, and political conditions to another very different set, the truly key features may be missed. For example, when asked why it is health workers will do the "right thing" even though there is no disciplining device of consumer choice, they are underpaid, there are no effective institutional controls, or no legal restraint like malpractice suits, the answer is often something like "they will behave because they are health professionals." Indeed, with professionals from well functioning systems, this is most likely the right answer. Doctors and nurses do not perceive themselves as performing for the money or because of threats, but out of professional pride and affiliation. The underlying factors of compensation and punishment are invisible not because they are weak or absent, but precisely because they are so strong and effective that gross deviations from appropriate behavior 'are rare, and hence the need to invoke explicit punishment similarly rare. However, when the underlying control mechanisms are weak or non-existent, professionalism is not likely to be a powerful enough inducement. Second, there are big advantages to having a mix of public and private with mobility between the two as a disciplining device on the behavior of individuals while in the public sector. Doctors working in a public sector hospital when there is a large and effective private 27 sector do not want to damage their reputation by performing noticeably worse than their colleagues. If there is no interchange either at the one extreme because the private sector is small, or at the other extreme the interchange is so large and fluid because health professionals work in both simultaneously and there is no standard for the public part of their practice, this regulating mechanism will cease to be effective. Lessons learned. What are the lessons from the positive experiences where the public sector was effective at improving health outcomes? Close examination of campaigns where PHC-like programs have appeared to be successful have pointed to the importance of social, political, and institutional ability to motivate performance from health workers. Two examples are to the point. First, Caldwell (1986) describes instances in Kerala, where community action held health workers accountable through strong-armed means: "Doctors and others who provide village services (for instance, bus drivers plying regular routes) know stories of their fellows who were treated violently or hurt in protests about their having failed their duty" (p. 199).21 The important role of social and political factors in generating the generally high performance of government services in Kerala is well described in Heller (1996). The second example comes from an assessment of a major health campaign in the Ceara state of Northeastern Brazil which contributed to a 36 percent fall in infant mortality in only a few years (Tendler and Freedheim, 1994). Through a careful examination of the details 2"Caldwell (1986) also quotes from a colleague working in West Bengal where success was achieved "because [the state's communist govermnent] have used the party system to appoint cadres at every health center to report on doctors or nurses who do not put all their time and effort into their services or who discriminate between patients" (p.203). 28 of the implementation of the health program, Tendler and Freedheim identify three primary reasons behind this success. First, through a merit hiring system and a large advertising campaign "the state succeeded in creating an sense of 'nmission' around the program and remarkable respect for its workers in the communities in which they served." Second, flexibility in job descriptions allowed workers to take on tasks that although "sometimes viewed as distractions by experts [they] formed the basis for relations of trust between workers and citizens." Third, job candidates, most of who were rejected, were educated about what to expect from workers, supervisors, and mayors, turning them into "informed public monitors of a new program in which the potential for abuse was high. "I Motivating public (and private) sector workers to deliver high quality services is a long-standing issue (and not just in the health sector), especially in situations where monitoring of the key elements of performance is difficult or costly. As the examples above point out, such motivation seems to have been achieved through a variety of means: direct monitoring with the threat of job loss, community monitoring with various threats, and community oversight and participation with the threat of job (and prestige) loss. Although the declaration at Alma Ata stated that "community participation" was a key feature of the PHC strategy, it appears to be more the exception than the rule. 22 For example job applicants, and many eavesdroppers who listened in on the hiring process, were told "Those of you who are not selected must make sure that those who are chosen abide by the rules. [...] If these rules are breached we want to hear about it. [...] we are keeping all the applications, just in case any of those we hire do not perforn well." (Tendler and Freedheim, 1994, pp 1777-1778) 29 IV) How does public sector provision affect the use of services by individuals? A second possible reason large scale provision of some types of PHC services would have little or no impact on health status is if the extension of public supply merely "crowds out" the consumption of nearly equally effective services from private sources. Even if health services are delivered effectively by the government, the health impact of these services depends not on the total use of public services, but rather how public provision affects total use. How big is the crowding out effect? A simple model of supply (S) and demand (D) will be useful to organize the evidence. The public sector is subscript "b," "v" denotes the private sector, and P refers to the price. Sb = Db S,(P,) = D,(PV,Sb) with total service use as D = Db + D,(P,) Alternatively, with a slight modification for the case when public and private services are essentially the same (perfect substitutes) in consumers' eyes: Sb + St(P,) = D(P,) If the public sector expands service provision by a given amount, dSb, (say in terms of capacity to see a certain number of patients per week), the total amount of added services used will be: 30 d D e + sX Cb Dv d Sb eS - D Sb where es refers to the elasticity of private services with respect to price, eD refers to the elasticity of demand for private services with respect to price, and ev,, is the elasticity of demand for private services with respect to the availability of public services (es> 0, ED < O, eVb2 times the minimum wage. The population distribution across these groups (in percent) is 11,17,23,22,27. Honduras: categories are poorest, intermediate, least poor whose distribution is 31,19,50 percent of households Nicaragua: categories are extreme poor, poor, non-poor, whose population distribution is 19.4, 30.9, 49.7 percent. See Appendix 3 for sources. 61 Appendix Table C-2: Distribution of the benefits of public spending (subsidies) on health by household income or consumption per capita quintile. Percent of benefits Country Year Type Poorest q2 q3 q4 Richest Argentina 1991 PHS+ 33 61 6 Brazil 1985 PHS+ 41 42 Bulgaria 1995 All health 3 1 21 26 6 Primary facilities 16 17 21 25 21 Hospitals 11 16 20 16 27 Chile 1982 PHS+ 22 7 7 7 67 7777 1 r Colombia 1992 Public health system 27 26 13 ISS (Inst. Soc. Sec.) 18 42 35 21 15 Ghana 1994 All health 12 1l5 19 21 3 Primary facilities 10 17 19 23 31 Hospital outpatient 13 15 17 19 35 Hospital inpatient 11 14 20 23 32 Guyana 1994 Public hospital 19 23 18 28 12 Public health center 28 16 21 22 14 Public health post 67 3 7 16 7 Public medicines 18 7 11 31 32 Indonesia 1987 Total 27TT 25T_ - 7T Hospital 8 11 17 28 35 Public health center 18 19 21 24 17 Jarnaica 1989/92 Hospital 19 18 23 23 18 Health center 25 30 16 IS II Kenya 1992 Total 14 27 22T 24 Hospitals 13 16 22 22 26 Public health center 24 23 23 17 13 Disp 17 20 23 19 19 Madagasar 11993/94 Public eath center i 12 14 3 3 Maiaysia T§18 PHS+ 29 60 11 Niciama 1199 Pre-natal and birffs . 12 31 57 J Curative care 10 28 63 Tanzania 19937 Hospital outpatient: curative 1 15 23 37 Hospital outpatient: prenatal 18 22 15 25 21 Hlth cntr outpatient: curative 18 21 19 21 21 Hith cntr outpatient: orenatal 25 15 21 18 21 Trinidad 1992 Hospital 17 23 25 16 19 and Tobago Health center 8 35 14 7 36 Souhfrie__ a 1993 PHS+ 16 07 17 St Lucia 1995 Public hospitals 26 28 14 13i Public health centers 25 26 15 11 24 Uruguay* 198 Total 377 17 14 11 Ministry of Public Health 57 25 12 5 1 Public hospital 44 22 15 12 7 Vietnarn _VP -93 All 12 16 2122 29 Commune health centers 19 29 24 19 10 Hospital outpatients 9 14 15 23 39 Hospital inpatients 13 17 25 22 24 Notes:* Persons are assigned to quintile on the basis of the ranking of households (i.e. the number of individuals in each quintile is non-constant). Nicaragua: categories are extreme poor, poor, non-poor, whose population distribution is 19.4, 30.9, 49.7 percent. + Source listed spending only as "Public health spending". See Appendix 3 for sources. 62 Sources fo' benefit incidence numbers: World Bank Poverty Assessments and country studies: Brazil, 1995; Colombia, 1994; Honduras, 1994; Kenya, 1995; Madagascar, 1996; Mongolia, 1996; Nicaragua, 1995; Tanzania, 1995; Uruguay, 1993; Vietnam, 1996 Madagascar Poverty Assessment 1996: Argentina, Brazil, 1985, Chile, South Africa, Malaysia Baker, Judy, 1997. Poverty Reduction and Human Capital Development in the Caribbean: A Cross-Country Study, The World Bank: Guyana, Jamaica, Trinidad and Tobago, St Lucia: Demery, Lionel, 1997, "Benefit Incidence Analysis" mimeo: Bulgaria, Ghana. Van de Walle, Dominique, 'The Distribution of Subsidies through Public Health Services in Indonesia, 1978-87," The World Bank Economic Review, 8(2):279-309: Indonesia. 63 Policy Research Working Paper Series Contact Title Author Date for paper WPS1856 Surviving Success: Policy Reform Susmita Dasgupta November 1997 S. Dasgupta and the Future of Industrial Hua Wang 32679 Pollution in China David Wheeler WPS1857 Leasing to Support Small Businesses Joselito Gailardo December 1997 R. Garner and Microenterprises 37664 WPS1858 Banking on the Poor? Branch Martin Ravallion December 1997 P. Sader Placement and Nonfarm Rural Quentin Wodon 33902 Development in Bangladesh WPS1 859 Lessons from Sao Paulo's Jorge Rebelo December 1997 A. Turner Metropolitan Busway Concessions Pedro Benvenuto 30933 Program WPS1 860 The Health Effects of Air Pollution Maureen L. Cropper December 1997 A. Maranon in Delhi, India Nathalie B. Simon 39074 Anna Alberini P. K. Sharma WPSI861 Infrastructure Project Finance and Mansoor Dailami December 1997 M. Dailami Capital Flows: A New Perspective Danny Leipziger 32130 WPS1 862 Spatial Poverty Traps? Jyotsna Jalan December 1997 P. Sader Martin Ravallion 33902 WPS1 863 Are the Poor Less Well-insured? Jyotsna Jalan December 1997 P. Sader Evidence on Vulnerability to Income Martin Ravallion 33902 Risk in Rural China WPS1 864 Child Mortality and Public Spending Deon Filmer December 1997 S. Fallon on Health: How Much Does Money Lant Pritchett 38009 Matter? WPS1865 Pension Reform in Latin America: Sri-Ram Aiyer December 1997 P. Lee Quick Fixes or Sustainable Reform? 37805 WPS1 866 Circumstance and Choice: The Role Martha de Melo December 1997 C. Bernardo of Initial Conditions and Policies in Cevdet Denizer 31148 Transition Economies Alan Gelb Stoyan Tenev WPS1867 Gender Disparity in South Asia: Deon Filmer January 1998 S. Fallon Comparisons Between and Within Elizabeth M. King 38009 Countries Lant Pritchett WPS1868 Government Support to Private Mansoor Dailami January 1998 M. Dailami Infrastructure projects in Emerging Michael Klein 32130 Markets Policy Research Working Paper Series Contact Title Author Date for paper WPS1869 Risk Reducation and Public Spending Shantayanan Devarajan January 1998 C. Bernardo Jeffrey S. Hammer 31148 WPS1870 The Evolution of Poverty and Raji Jayaraman January 1998 P. Lanjouw Inequality in Indian Villages Peter Lanjouw 34529 WPS1871 Just How Big Is Giobal Production Alexander J. Yeats January 1998 L. Tabada Sharing? 36896 WPS1872 How Integration into the Central Ferdinand Bakoup January 1998 L. Tabada African Economic and Monetary David Tarr 36896 Community Affects Cameroon's Economy: General Equilibrium Estimates WPS1873 Wage Misalignment in CFA Countries: Martin Rama January 1998 S. Fallon Are Labor Market Policies to Blame? 38009