POLICY RESEARCH WORKING PAPER 2769 Immunization in Developing Countries Its Political and Organizational Determinants Varun Gauri Peyvand Khalegbian The World Bank Development Research Group Public Services January 2002 I POLICY RESEARCH WORKING PAPER 2769 Abstract Gauri and Khaleghian use cross-national social, political, bureaucratic elites have an affinity for immunization economic, and institutional data to explain why some programs and are granted more autonomy in autocracies, countries have stronger immunization programs than although this effect is not visible in low-income others, as measured by diptheria-tetanus-pertussis (DTP) countries. The authors also find that the quality of a and measles vaccine coverage rates and the adoption of nation's institutions and its level of development are the hepatitis B vaccine. After reviewing the existing strongly related to immunization rate coverage and literature on demand- and supply-side factors that affect vaccine adoption, and that coverage rates are in general immunization programs, the authors find that the more a function of supply-side than demand effects. elements that most affect immunization programs in low- There is no evidence that epidemics or polio eradication and middle-income countries involve broad changes in campaigns affect immunization rates one way or another, the global policy environment and contact with or that average immunization rates increase following international agencies. Democracies tend to have lower outbreaks of diphtheria, pertussis, or measles. coverage rates than autocracies, perhaps because This paper-a product of Public Services, Development Research Group-is part of a larger effort in the group to study the political economy of basic service delivery. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Hedy Sladovich, mail stop MC3-3 11, telephone 202-473-7698, fax 202-522- 1154, email address hsladovich@worldbank.org. Policy Research Working Papers are also posted on the Web at http:i/ econ.worldbank.org. The authors may be contacted at vgauri@worldbank.org or pkhalegh@Cjhsph.edu. January 2002. (42 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the 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 view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Research Advisory Staff Immunization in Developing Countries: Its Political and Organizational Determinants Varun Gauri Development Research Group, The World Bank, Washington, D.C. Vgauri(worldbank.org Peyvand Khaleghian Johns Hopkins University pkhalegh@jhsph.edu Introduction Why do some countries consistently immunize children more effectively than others? Both practical and theoretical concerns motivate the question. Practically, there is little systematic evidence about what policies and vaccination strategies expand immunization coverage. Understanding the relative importance of campaigns versus routine programs, the role of international donors, access to health care facilities, the availability of communication equipment, and other factors will help health policy planners design immunization programs. In addition, a number of foundations, multilateral agencies, governments, and analysts have come to believe that in developing countries vaccines will be necessary to halt the spread of devastating diseases, such as HIV/AIDS and malaria, and that developing countries and their governments, left to themselves, will not buy enough vaccines to stimulate market development of vaccines for those diseases. This paper examines the question of why some developing country governments do not purchase and deliver enough vaccines to their populations, and what might be done about this, both for existing and future vaccines. The theoretical motivation involves the political sources of social policy. For some policy outcomes, hyper-rational accounts of government choices are not tenable, and political explanations are unavoidable. In the case of vaccines, every society recognizes that immunization is a task for government, the materials cost of the basic six childhood -1I- vaccines is low (less than $1 per filly immunized child), and financial and technical assistance for vaccination is available from international organizations. Despite this, many governments fail to immunize their populations adequately. Related theoretical problems appear in other sectors. Many governments adopt ruinous economic and education policies despite a mountain of evidence regarding their irrationality. Generally, analysts explain these policy outcomes by identifying winners and losers, assessing the political, economic, and cultural resources available to various interest groups, and examining patterns of decisionmaking institutionalized in customs and laws. This paper uses a panel data set of immunization coverage rates and the timing of vaccine adoption in low- and middle-income countries, along with a time series of potential political, economic, social, and institutional determinants, to identify the political and organizational factors associated with strong immunization programs. The debate over how countries have improved health outcomes has two main camps. One camp emphasizes the role of economic growth and related gains in nutrition (McKeown 1976, Fogel 1984); the other camp, who are writers, point to the importance of public health measures such as communicable disease control and skilled attendance at birth. (Preston 1976). Each group has favored country examples: for the former the United States and Britain and Wales, where public health was weak when mortality rates fell sharply in the nineteenth century, and for the latter Sri Lanka and the Indian state of Kerala, where mortality rates fell in this century despite extremely modest gains in national income. In fact, there is nothing inherently incompatible about the views espoused by the two camps. Public health measures reduce the likelihood of individuals being exposed to pathogens, whereas income and nutrition contribute to their ability to resist or recover from exposure. Achieving a life expectancy over 70 probably requires both a minimal income and nutritional level, as well as a reasonably strong public health system, but at lower life expectancies the two factors might function as substitutes. (Johansson and Mosk 1987) The politics, economics, and history of any given country might make either the income or the public health path to low mortality easier to achieve. This paper contributes to an emergent literature within the public health camp on the political, economic, and social characteristics of countries that are able to achieve low mortality despite low levels of income. That discussion focuses on, to put it simply, how democracy and social mobilization can contribute to the institutionalization of a public health agenda, as in Kerala, Sri Lanka, and Costa Rica, and how hierarchy and revolutionary ideology can create the same, as in Cuba, Vietnam, and China. -2- Kerala, Sri Lanka, and Costa Rica are autonomous regions whose achievements in raising life expectancy and lowering infant mortality are widely cited. Those societies share several characteristics: small and densely settled populations, a relatively high degree of female autonomy, a respect for education, and the absence of a rigid class structure. At the same time, their governments have historically allocated significant resources to the health care system, and the political histories of the regions are characterized by egalitarianism and a national consensus with marked elements of populism that emerged from political contestation. (Caldwell 1986) Their histories are consistent with theoretical treatments of the impact of democracy on social policy: competitive elections motivate politicians to provide social services to capture the support of the poor, the poor face higher costs in mobilizing for social services in authoritarian governments, and democracies empower interest groups and social movements advocating propoor social policies. (Lake and Baum 2001, McGuire 2001, Weyland 1996) But Vietnam, China, and Cuba are also countries that have improved the health of their citizens dramatically despite relatively low levels of income per capita. In those countries political contestation played no role, but disciplined party organizations structured in a quasi-military fashion and social revolutions followed by high levels of government expenditure on health and social infrastructure did (Bryant 1998, Parish and Whyte 1978). This paper assesses the impact of competitive elections on one clear and measurable facet of public health outcomes, immunization. Immunization programs differ in some ways from most other kinds of health care. Consensus among technical experts in the field of immunization is stronger, the time-frame for policy reforms is shorter (if "reform" is even the right word for increasing coverage rates or adopting a new vaccine), household demand for it is weaker than for curative health care, and the professional and pecuniary interests of the providers are weaker in companson to other areas of health reform, such as changes in insurance and provider payment systems. In addition, multinational pharmaceutical corporations operating in an oligopolistic market and the international scientific community have unusually strong influence on the vaccines distributed in developing countries. So the explanation of immunization outcomes in this paper might not be co-extensive with a broader theoretical account of public health programs in developing countries. In particular, there are reasons to believe that the political and organizational determinants of immunization programs might depend less for their effect on the way they either promote or dampen the demands of interest groups and households, which is how institutions are understood to operate in most political economy accounts of social service provision, and more on their direct effect on the actions of political and bureaucratic elites. -3- There has been little work that disaggregates the categories of public health and examines the political and organizational sources of policies in its subsectors. One noteworthy exception is an historical treatment by Nathanson (1996), whose arguments are suggestive for a comprehensive account of the political economy of public health, which will require case studies in addition to cross-national approaches like that taken in this paper. In the late nineteenth and early part of the twentieth centuries France enacted national legislation for the inspection of wet nurses, paid maternity leave, free medical care during confinement, and obligatory maternity insurance, while the United States lent no federal attention to maternal and child health until maternal health education programs began in 1921 (only to be rescinded in 1929). On the other hand, beginning in 1965 the United States mandated warnings for tobacco consumers, curtailed tobacco advertising, banned smoking during interstate transportation, and raised cigarette excise taxes, which jointly had the effect of lowering U.S. cigarette consumption sharply; but France has not reduced cigarette consumption and has achieved only limited compliance with smoking bans in public facilities. Nathanson uses two main factors to explain why maternal and child health concerns appear unrelated to antismoking efforts in the two countries. First, the French state is more centralized than the American, so resources for service provision are more easily mobilized when there is state consensus. Since improving maternal and child health requires additional resources, it is not surprising that France has a more active record than the United States in that area. Centralization also explains French weakness in antismoking efforts. Alternative channels for social policy change, such as local governments and social movements, are not available when the French state is divided, which it happens to be in this area because government-owned tobacco producers resist controls. In a decentralized country like the United States, alternative government entities, such as state court and legislatures, are available when action at the federal level is blocked, and it is in fact at the state level where the antismoking campaign first gained its momentum. Nathanson's second explanatory factor is the style of risk construction. United States social and political culture tends to construct health risks as dangers to individuals, whereas the French style of risk constructions emphasizes threats to the nation. Consequently, the rights of individual nonsmokers were dispositive for legislative and court action in the United States, and the language of French nationalism was critical for the pronatalism that helped to justify maternal and child health policies in France in the 19th century. Prevailing Views on the Determinants of Immunization Coverage Most studies use household surveys and explain variance in childhood immunization "uptake" or "demand" with characteristics of children's mothers and their - 4- households, often but not always including community-level fixed effects. These studies invariably find maternal education and household socioeconomic to be correlated with the probability of childhood immunization, but there are disparate findings concerning the extent to which these are causally related to immunization status, and whether measurable maternal and household characteristics might be proxies for other underlying factors or for characteristics of the communities of residence. For instance, Desai and Alva (1998) use the first round of data from the Demographic and Health Surveys (DHS) in 22 countries and find that whereas the inclusion of individual-level and community-level fixed effects significantly weakens the relationship between maternal education and childhood health, the link between maternal education and child immunization remains strong. In a survey in two villages near Yogyakarta, Indonesia, Streatfield, Singarimbun, and Diamond (1990) also find that immunization status is related to maternal education, albeit in a U-shaped pattern in which illiterate mothers are more likely to have their children immunized than mothers with a little primary education, but much less likely than mothers with some secondary schooling. They explain the U-shaped pattern by arguing that illiterate mothers comply more often with social norms, but that higher levels of education are associated with knowledge of the functions of the vaccines, which has a strong and independent effect. Their multivariate analysis finds that the effect of formal education disappears when mothers have correct knowledge of the functions of vaccines. In another study, Gage, Sommerfelt, and Piani (1997) find that higher household socioeconomic levels and more maternal education both increase the likelihood of childhood immunization in Nigeria and Niger, and that household structure (living in a nuclear or elementary polygynous family, as opposed to a laterally extended family) negatively affects immunization probabilities in Nigeria, but not in Niger. In a study of family choices for maternal and child health in Guatemala, Pebley, Goldman, and Rodriguez (1996) find that both mother's and father's education are significantly and positively related to childhood immunization status, as is living in urban areas, but that unobserved family and community characteristics are even more influential. They hypothesize that family health beliefs, differing abilities among families to take advantage of available resources, and variance in the intensity of immunization campaigns in different areas might explain these intraclass correlations. In a study of immunization uptake in four rural areas of Bangladesh, Steele, Diamond, and Amin (1996) find that the effect of mother's education on child immunization status disappears once father's education is included, and that the latter became insignificant when village-level dummy variables are added. They find a large amount of unexplained variation at both the household and village levels, which they speculate might be related to immunization accessibility, the attitudes of local leaders, differences in household attitudes and beliefs, and power relationships within -5- the household. Their study offers evidence for the power relationship interpretation since children who lived with their mothers and patemnal grandparents were significantly less likely to be immunized, and children whose mothers belonged to women's social groups were significantly more likely to be immunized. Anthropological accounts of the demand for or acceptance of immunization center on the relationship between modem germ theory and local beliefs. For example, villagers in South Asia have been reluctant to accept vaccines for smallpox and measles out of a belief that a goddess causes those diseases, and that treating a body that she has inhabited will anger her and only make the disease worse (Caldwell, Reddy, and Caldwell 1983, Nichter 1995). Still others do accept immunization, particularly for smallpox, because they interpret the vaccination mark as a kind of talisman that can ward off the spirit or the goddess. The Sri Lankan Ministry of Health has used this association by placing an amulet on a national vaccination poster. The conflation with talismans can be misleading, however, because these, unlike childhood immunizations, are believed to reduce the severity of afflictions without eliminating their occurrence, and to have a temporary effect. Other reasons for refusing vaccines include the impression that if a vaccinated child is healthy she does not need any more immunizations, the idea that a vaccination protects against all illness and the resulting disappointment if a vaccinated child falls ill, the fear that the side effect from a given vaccine will also occur following all vaccines, the fearful association between the syringe and the Christian cross, and the impression that immunization is related to sterilization efforts or medical experimentation (Nichter 1995). These beliefs make clear how education might work to increase immunization acceptance, not only by providing information, endowing the skills to acquire further information, and strengthening the status of certain household members, but also by creating a sense of identification with the modernity and its associated schools, clinics, and practices (Caldwell 1983, 1986). It is important to note that the fear of immunization is not a monopoly of developing countries. Classic accounts include Voltaire's mocking portrayal in 1733 of his own countrymen for their suspicion of the British use of smallpox vaccination: "It is inadvertently affirm'd in the Christian Countries of Europe, that the English are Fools and Madmen. Fools, because they give their Children the Small-Pox to prevent their catching it; and Mad-men, because they wantonly communicate a certain and dreadful Distemper to their Children, merely to prevent an uncertain Evil." (Letters concerning the English nation, 1733, "Letter XI on Inoculation"). In his Autobiography, Benjamin Franklin noted, and regretted, his own similar estimation of the value of inoculation: "In 1736 I lost one of my sons, a fine boy of four years old, by the small-pox, taken in the common way. I long regretted bitterly, and still regret that I had not given it to him by inoculation. This I - 6 - mention for the sake of parents who omit that operation, on the supposition that they should never forgive themselves if a child died under it; my example showing that the regret may be the same either way, and that, therefore, the safer should be chosen." An influential pamphlet circulated in Sweden in 1818 argued that vaccination would anger God and make the disease worse (Skold 1996). A modem form of this argument, in which people fear that immunization can damage health by driving a disease underground, continues to recur in the United States and elsewhere (James 1988), as do persisting concerns that vaccines may increase the likelihood of diseases such as autism, in spite of considerable evidence to the contrary (Taylor and others 1997). These fears of immunization have had documented negative effects on vaccine coverage rates. Gangarosa and others (1998) reviewed antivaccine movements targeted against the pertussis vaccine in 12 Western countries and found a strong correlation between the timing and strength of the movements and declines in vaccine coverage rates. For example a 1974 report in the United Kingdom ascribed 36 neurological reactions to the whole-cell pertussis vaccine, and although health authorities resisted pressure to withdraw the vaccine during the panic that followed, the DTP coverage rate fell from 81 percent in 1974 to 31 percent after 1976, and a spike in pertussis incidence ensued. When an influential medical leader in Sweden claimed that pertussis had become milder as a result of economic, social, and medical progress and questioned the value of the vaccine, and when Swedish pediatricians lost confidence in the vaccine as a result of those questions and the attribution of some neurological events to the vaccine, the coverage rate plummeted from 90 percent in 1974 to 12 percent in 1979. The pertussis incidence rate in Sweden increased thirtyfold, with rates of serious complications approaching global rates. Potential distrust of immunization is perhaps unavoidable because vaccinations entail, on the face of it, an inversion of the healing paradigm: one goes when healthy to the clinician, who injects a substance in one's arm that causes discomfort and, in rare occasions, an actual case of illness. But it is also true that misleading or misunderstood information, expired and ineffective vaccines, dirty or reused syringes, and poorly trained staff can undermine trust in immunization programs, resulting in lower coverage rates. Supply-side factors that affect immunization coverage, such as the quality of vaccines, incentives for the cadre of workers who perform vaccination, and the organization of national immunization agencies, have received little attention in the literature. One study in Gujarat state in India argues that the agents who perform vaccines for the Indian public sector, the so-called multi-purpose workers who live in and work from their home villages, were excessively focused on numerical targets, so much so that they did not explain the purposes, benefits, and potential side effects of vaccines to patients, nor did they convey - 7 - potentially useful information up to their superiors regarding the obstacles to immunization, such as geographic challenges, caste and gender discrimination, and the influence of mothers-in-law (Streefland 1995). Several accounts of immunization policy have used the concepts of "political will" or "political commitment" to explain the success of moves to improve coverage (UNICEF 1996 pp. 65-66, Madrid 1998a, Widdus 1999a, Justice 2000b, 2000c), conduct polio eradication activities (Hull and Aylward 2001), or introduce new vaccines (Madrid 1998b, Huang and Lin 2000, Miller and Flanders 2000, Wenger 2001). Put simply, the implication is that if political leaders were to make immunization a priority, coverage rates (or polio eradication, or the introduction of new vaccines) would improve as a result. However, as several writers have pointed out, the terms "political will" and "political commitment" are "catch-all culprits" without much analytic content (Grindle and Thomas 1991, Reich 1994). As one commentator put it, "political will [does] not arrive de novo, so it is necessary to understand the elements that contribute to recognition of need, and willingness (or capacity) to pay [for new vaccines]" (Widdus 1 999a). Some of these case studies have also pointed to the negative effects of decentralization on immunization rates. (Justice 2000a,b and Madrid 1998a,b). In theory, decentralization might make coordination among local health secretariats more difficult, and it might create a free rider effect in which the incentives for one jurisdiction to immunize its population are negatively related to the strength of the immunization effort in neighboring jurisdictions. But it might also, under certain conditions, improve service delivery by making governments more accountable to and responsive to needs of local populations. Studies that examine the determinants of vaccine adoption have focused primarily on the informational prerequisites for government decisionmaking. Mahoney and Maynard (1999), reviewing the experience of the International Task Force on Hepatitis B Immunization, found five factors to be significant: (1) the establishment and dissemination of disease burden data and cost-effectiveness computations; (2) vaccine introduction trials and effectiveness evaluations; (3) establishment of an international consensus on recommendations for vaccine use; (4) assurance of an adequate and competitive vaccine supply; and (5) the creation of funding mechanisms to supply vaccine to countries unable to finance their own procurement. Hausdorff (1996), commenting on the rate of new vaccine adoption, suggested that the main determinant is "the extent to which international and bilateral agencies and national governments appreciate the potential value of new vaccines." A study by the General Accounting Office of the United States (1999) drew similar conclusions regarding the importance of locally tailored information, both for new vaccine adoption and for decisions to invest in immunization programs as a whole, and highlighted the role of disease surveillance. Levine and Levine (1997) emphasized the role of information, public perceptions, and the opinions of the medical community, as did - 8 - Wenger and others (2000) in their analysis of H. influenzae type B vaccine adoption. Wenger (2001), Hausdorff (1996) and GAO (1999) pointed to the importance of price barriers. Widdus (1999b) analyzed both prerequisites (recognition and information on the target disease, the presence of a functioning immunization program) and modifying factors (epidemiologic and demographic variables, cues for action) that jointly shape a "perceived threat of disease" which, in turn, influences decisionmakers' perceptions of the risks, benefits and barriers associated with new vaccine adoption. Miller and Flanders (2000) identified a number of epidemiologic and economic factors associated with the uptake of hepatitis B and Haemophilus influenzae type B vaccines into national immunization programs. In their model, variables significantly affecting the odds of adopting the vaccine included GDP per capita (with an odds ratio of 4.4), vaccine cost per dose (OR 4.1), vaccine cost as a fraction of per capita GDP (OR 39.7), current coverage with DTP vaccine (OR 55.1), and years of life lost per 1,000 infants due to the diseases in question (OR 6.9). Using alternative models with composite variables, they find that a variable for potential treatment costs prevented-a composite of data on disease burden, unit treatment costs and immunization coverage rates-has the highest odds ratio, at 127.4. All of these studies, while pointing to the importance of information prerequisites leave unexamined the question of what motivated decisionmakers to conduct epidemiological analyses to begin with, the question of, in other words, the political sources of immunization policy. Where they are consistent with the present analysis is the finding that the role of bureaucratic elites has been important in shaping immunization policies over the past two decades. The history of smallpox vaccination offers lessons regarding the organization of immunization efforts. The Swedish Crown encountered significant resistance when it tried to implement a campaign for smallpox inoculation in the eighteenth century, even though some 300,000 people died of smallpox between 1750 and 1800; but its subsequent campaign using the vaccine derived from cowpox virus was much more successful. Sweden became one of the first countries to make vaccination compulsory in 1816 and vaccinated 80 percent of newborn children by 1821. Skold (1996) attributes this to the superiority of vaccination technology, but also to institutional factors, such as the influence of the international press (80 percent of articles published between 1798 and 1801 favored vaccination), the removal of the authority to vaccinate from a physicians' monopoly, the enlistment of the clergy and church assistants, who did most of the vaccinating and received certificates of competence from the district medical officer, free vaccines offered to the poor, and a system of rewards given to the "most skilled" vaccinators. Rigau-Perez' (1989) study of the introduction of smallpox inoculations to Puerto Rico in 1803 also emphasizes the critical role of the clergy in obtaining the community's trust. In Puerto Rico the colonial authorities had anticipated the Royal Expedition of the Vaccine from Spain and introduced -g9- inoculation techniques with so much community involvement "that children, in their games at school, vaccinated each other." But the initial success of the effort foundered as a result of conflicts between colonial and royal authorities, and Puerto Rico's smallpox program lagged far behind many of its Latin American neighbors, lending an ominously prophetic tone to a comment of the King's emissary, who had said in a letter that he was coming "not only to bring vaccine, but to assure its perpetuation, which is the hardest [task]." Variables and Sources of Data Immunization coverage. Data on immunization coverage rates obtained from WHO and UNICEF measure the proportion of children who have received the DTP3 or measles vaccine by one year of age and are based on either service delivery records or, where available, on coverage surveys carried out under the auspices of the EPI program using a standardized 30-cluster sampling technique which collects information directly from households (Henderson and Sundaresan 1982, WHO 1991). l Coverage rates are recorded without indication of the specific methods used to collect them, though survey data-which are considered to be more valid than service statistics, particularly in countries where routine reporting systems are not well-developed-are used wherever possible, including from sources outside the EPI system such as the immunization section of the Demographic and Health Surveys (UNICEF 1996). Several studies have voiced concern over the disparity between coverage rates reported from service delivery records and "actual" coverage rates measured by sample surveys (Boerma and others 1990, UNICEF 1996, WHO 1999). In one study, 30.4 percent of reported figures were considered "unexpected" when assessed against a set of internal and external validation criteria (WHO 1999); in another, reported coverage was found to exceed survey coverage by over 20 percent in a small sample of case-study countries including Uganda, India, and the Philippines (UNICEF 1996). Various explanations are possible for these discrepancies, which may derive from inaccuracies in either reported or survey data. Survey data, while generally considered more valid than data from service records, are not invariably accurate. Case definitions and the method of obtaining data may differ from survey to survey, and other differences in survey methodology-between the EPI's cluster sampling method and the Demographic and Health Surveys (DHS) approach, for example-may also result in different estimates for the same population (Boerma and others 1990). Surveys typically obtain coverage information from maternal recall, immunization cards or both (WHO 1991, Boerma and others 1990, Boerma and Bicego 'While immunization schedules vary from country to country, most require the DTP3 vaccine to be given at 3 to 6 months and the measles vaccine at 9 to 12 months. For more details, see WHO (2000). - 10 - 1993). "Card plus history" approaches are generally the most accurate, though even these underestimate "true" coverage levels to a small extent. Some surveys use a sampling frame that is limited to "women of child-bearing age," generally 15 to 49 years, which may additionally underestimate true coverage levels by excluding children in the care of older caregivers-a common occurrence in countries suffering from conflict or, more recently, HIV/AIDS (World Bank 1997). Other methodologic issues have also been discussed, with criticisms of the ±10 percentage point accuracy of the WHO's cluster sampling method (Lemeshow and others 1985) and discussion of other methods of obtaining coverage information such as modified cluster surveys and Lot Quality Assurance methods (Bennett and others 1991; Lemeshow and Robinson 1985; Lanata and others 1990; Sandiford 1993; Turner, Magnani, and Shuaib 1996; WHO 1996a; WHO 1999; Hoshaw-Woodard 2001). However, given the simplicity and inexpensiveness of the WHO's 30-cluster survey approach, and the fact that it generally compares well with data from nationally representative surveys such as the DHS (Boerma and others 1990, Boerma and Bicego 1993), this method remains the most common way of obtaining coverage data by survey. Coverage data from service records are also subject to a wide range of inaccuracies. Like survey data, reported data can suffer from nonsystematic errors such as mistakes in data entry, transcription and analysis. Also, since service records are seldom complete, national coverage levels must be estimated from available data using a variety of assumptions, introducing an additional source of possible error (Boerma and others 1990). Children over the age of one are frequently included in the numerator of reported data, a systematic error that artificially inflates rates and gives an inaccurate picture of true coverage,2 while data on the target population-the denominator in coverage calculations- are frequently based on projections from old or inaccurate censuses and may seriously over- or under-state the correct denominator in these calculations (UNICEF 1996). Perhaps more serious, however, is the possibility of deliberate rate inflation or data falsification. This is of particular concern in systems where data is collected principally to measure progress against targets rather than to guide program decisions at local or district levels (UNICEF 1996). In such systems, pressure from superiors to reach unrealistic targets or increases in coverage may lead health workers to inflate coverage data, and these in turn may find their way into national and international health statistics, particularly in the absence of effective monitoring and supervision systems.3 A separate possibility is that overworked front-line 2 Since the standard definition of coverage is based on a child's vaccination status at one year of age, -including children who have received these vaccines after their first year artificially increases the numerator in coverage calculations. 3 These pressures can originate ftom intemational organizations as well as national public health bureaucracies, such as during the last phases of the "Universal Child Immunization" initiative when UNICEF - II - health workers may neglect immunization responsibilities-especially if other programs come with stronger incentives or are perceived as having greater priority-and may attempt to conceal this by fabricating, falsifying or exaggerating data (UNICEF 1996, Nair and others 2001 ).4 The key question for the present study is the extent to which these considerations may affect the study's analysis. Setting aside random errors, the principal concern is whether there are systematic errors in the coverage data that differ according to other country characteristics such as national income. To examine this, reported data on measles and DPT3 coverage from the WHO database were compared with similar data from DHS surveys for 82 country-years (DHS, 1985 to 1997). Figures 1 and 2 indicate the relationship between WHO and DHS data for these countries, for DTP3 and measles, respectively. There was no significant difference in the average measles coverage data reported in the DHS and WHO datasets, but DPT3 coverage was significantly higher in the WHO data by an average of 4.1 percent (s=2.04, p<0.05).5 More importantly, this difference in DPT3 rates was the same across all country income categories for which DHS data were available (low, lower middle and upper middle), suggesting that discrepancies between reported and "actual" rates do not differ according to countries' national income.6 Since DHS data were not available for any high-income countries, a possible difference there can not be ruled out; but in general, these findings suggest that using reported rates-particularly for measles7-as a dependent variable is reasonable, particularly given the absence of any itself admitted that "global pressure for achieving UCI" may have adversely affected the accuracy of reported coverage rates in 1990 (UNICEF 1996, page 40). 4 Nair and others (2001) report performance exaggeration in all programs other than imnunization in their study of village health workers in Kerala, India. They acknowledge that this might have been an artefact of programmatic factors specific to the Kerala context-in particular, the ability of households to obtain immunization from a wide variety of sources other than village health workers-rather than a demonstration of more accurate reporting of inmmunization figures by the VHWs themselves. 5 Figures I and 2 illustrate these errors as the distance of each country from the line yx. India, Yemen and Bangladesh, for example, report highly exaggerated coverage rates, up to 40 percent higher than the rates found in DHS surveys; while countries such as Kenya and Ghana report rates that are between 20 and 40 percent lower than those reported in DHS surveys. 6 Improbable reports of immunization coverage are also seen in some high income countries. Austria, for example, reports 90 percent coverage with DPT3 in every year since 1980, and reports its measles coverage in strict multiples of five. Apparent inaccuracies in coverage data may also be a result of transcription or data entry errors, such as New Zealand's 114 percent coverage rate for measles in 1991. This was manually corrected to 91 percent, based on its linear relationship with DTP3. ' It is interesting to speculate as to why reported measles figures tend to be more accurate than reported figures for DPT3. Since DPT3 requires the successful, timely completion of a course of three vaccinations, the cumulative likelihood of error-both in administering the vaccine and in recording or recalling this event- may be higher than for measles, which only requires a single shot and can also be administered in "catch-up" or "one shot" campaigns. This is borne out by coverage comparisons, in which measles coverage is generally higher than DPT3 coverage. (Boerna et al. 1996) This in itself doesn't explain differences in the accuracy of reported data for the two vaccines, however. It is possible that the greater discrepancy between reported and - 12 - significant evidence for systematic bias. Further attempts to explain intercountry differences in DPT3 figures between WHO and DHS data found only one variable, the log of total population, to be a significant determinant of these differences (p=O.03). 8 Figure 1: WHO vs. DHS Coverage Rates, DTP3 Vaccine 100 - Banglade Jordan y/ietnam Pej pt Jordan 90 - Yemen, dia lndone onBirlade IIe Rwanda Indonesi N amibia EgMfocco Morocco Kazakhst Douv iaj EgyEllivia TaF awi 80 - Ng a1ald a~~~~~~~~Comoros Turke Triank Pakisthdagasc Zimbabwe Y 70- Burkina GuaterBa l.Peru Trinidad a) NbIndi Brazil 0L a 60 - ParaguaioBzambiq C