The distribution and impact of health resource availability in China

World Bank Reprint Series: Number 397 Nicholas Prescott and Dean T. Jamison ne iiistrlbution and hnpact of Health Resource Availability in C ea Reproduced by permission of John Wiley and Sons Limited, England, from the In!tertifftiotalJournal of Healthl Planninig and Aftveiti,men,e,. vol. I (1985), pp. 45-56.  INTERNATIONAL JOURNAL OF HEALTH PLANNING AND MANAGEMENT, VOL. 1, 45-56 (1985) THE DISTRIBUTION AND IMPACT OF HEALTH RESOURCE AVAILABILITY IN CHINA* NICHOLAS PRESCOTr AND DEAN T. JAMISON The World Bank, 1818 H Street, NW, Washington, DC 20433, USA SUMMARY Since 1949, China's progress in mortality reduction has far exceeded that experienced by other developing countries with comparable levels of national income. This achievement has taken place irn the context of a development strategy oriented, in part, to the elimination of the worst aspects of poverty. Using recent cross-section data, this paper provides a statistical assessment of the extent to which health resources are evenly distributed in contemporary China, and the degree to which improvements in health resource availability may account for the observed variation in mortality levels. Contrary to expectation, the analysis finds that substantial inequalities do remain in the distribution of health resources, and that these differentials are principally associated with levels of urban income and urbanization. However, these differen.zes in health resource availabil- ity do not appear to explain the significant variation which also persists in mortality levels, a finding consistent with the results of similar analyses for developed countries. KEY WORDS: China; Mortality; Hospital beds; Health manpower; Regression analysis INTRODUCTION Before 1949, China's population suffered a crippling burden of disease and premature death. Perhaps the most striking success of China's subsequent struggle against poverty and inequality has been a dramatic increase in life expectancy, accompanied by a reduction in the burdein of illness. Life expectan- cy has more than doubled from only about 32 years in 1950 to 69 years in 1982-some twenty years greater than the average for other low-income countries, and only six years less than in the industrialized market economies. The successes associated with modern Chinese health policy have attracted considerable attention throughout the developing world. Indeed, China's policy emphases on prevention, barefoot doctors, and community resource mobiliza- tion strongly influenced the 1978 Alma-Ata declaration on achieving Health for All by the Year 2000 through a strategy of primary health care. The implementa- tion of these policies has also been accompanied by a rapid growth in the availability of conventional health resources, as measured in terms of hospital beds and health manpower. Since 1952 the availability of hospital beds has increased sevenfold from 0.28 to 2.07 per 1000, while that of doctors has nearly doubled from 0.74 to 1.32 per 1000 (State Statistical Bureau, 1984). *The views arid interpretations in this paper are those of the authors and should not be attributed to The World Bank, to its affiliated organizations or to any individual acting on their behalf. 0749-6753/85/010045-12$01.20 (C 1985 by John Wiley & Sons, Ltd.  46 N. PRESCOIT AND D. T. JAMISON These developments in policy, resource availability and mortality reduction provoke important practical questions about the lessons that may be learned from China's experience to help guide other developing countries which have set ambitious goals for health improvement. To what extent has China equalized the distribution of health resource availability? And to what extent can these improvements in resource availability account for the dramatic gains in life expectancy which China has achieved? In this paper we provide a statistical assessment of evidence relevant to answering these questions. Related papers assess factors influencing fertility levels in China (Birdsall and Jamison, 1984) and nutritional status (Jamison et al., in preparation). Extensive background material and data sources for the Chinese health sector are provided in Jamison et al. (1984) and Prescott et al. (1983). Most previous literatLQre attempting to assess the impact of aggregate mea- sures of health resource availability has examined cross-country variation. To take an example that used countries as the unit of analysis, and data from 131 member states of the World Health Organization, Fulop and Reinke (1981) found that, controlling for a country's income level and other characteristics, an increase in the number of physicians per capita was statistically significantly associated with higher levels of life expectancy. The magnitude of the effect was, however, small; they estimated that a 35 per cent increase in the number of physicians per capita would be required to effect a one-year gain in a country's life expectancy. Auster et al. (1969) condtii. 'ed an early study of this genre; using data on individual states of the USA they tound availability of medical services to have a beneficial (though weak) effect on mortality rates. Higher average levels of education, on the other hand, had a strong beneficial effect on mortality. Mendelson and Orcutt (1979) have used similar methods to assess the impact of air pollution on mortality; they showed that specific pollutants can be clearly related to mortality of particular types in particular age ranges. Cochrane and her co-workers (Cochrane, 1980; and Cochrane et al., 1982) have reviewed many analyses, concluding that improved education levels lead to reduced mortality and improved health. Paradoxically, the evidence relating health service availability to improvements in mortality reveals a mixed picture. In contrast to Fulop and Reinke, Cochrane et al. (1978) and Newhouse and Friedlander (1980), to take two examples, found no beneficial relation between niedical inputs and mortality or physiological measures of health status, respec- tively. The reviews of Cochrane (1980) and Preston (1976) concerning the effect of income levels on mortality also showed mixed results. In the sections that follow in this paper, we utilize similar methods to examine the case of China and extend previous research by assessing determinants of the distribution of resources. We begin with a discussion of data and methods; then follow sections presenting the main results. A final section draws conclusions. DATA AND METHODS The empirical basis for our analysis is provided by cross-sectional data available for the 29 provincial-level administrative units in China (provinces, autonomous regions and municipalities). These data are relatively recent, relating in most  HEALTH RESOURCE AVAILABILITY IN CHINA 47 cases to the years 1981 or 1979, and for the most part are available for all 29 provincial units. The definitions for the variables used are presented in Table 1. Our measures of health resource availability are both physical and financial. Data on hospital facilities are for total hospital beds of all types; general hospital beds (excluding hospitals of traditional medicine, hospitals affiliated with medical colleges, mental hospitals, tuberculosis hospitals, commune health centers and certain other categories); and commune health center beds only. Data on health manpower are for all salaried health workers; all doctors (including senior plus assistant doctors of Western and traditional medicine, but excluding barefoot doctors); and barefoot doctors only. With the exception of commune health centers and barefoot doctors, each of these physical resources does, in principle, serve the entire population of each province, so that its Table 1. Provincial variables: Definitions HOS81 Total hospital beds per 1000 population, 1981 H0S79 Total hospital beds per 1000 population, 1979 GENHOS81 General hospital beds per 1000 population, 1981 SHW81 Salaried health workers per 1000 population, 1981 SHW79 Salaried health workers per 1000 population, 1979 DOC81 Total doctors (senior plus assistant Western and traditional) per 1000 population, 1981 CHC81 Commune health center beds per 1000 rural population, 1981 BFD81 Barefoot doctors per 1000 rural population. 1981 REXP81 Public recurrent expenditure on health, 1981 (yuan)'$ CEXP8i Public capital expenditure on health. 1981 (yuan)* LEXP75 Life expectancy at birth, 1973-75 (years) CDR81 Crude death rate, 1981 (per 1000) A UT Dummy variable indicating autonomous region or not ILLIT82 Per cent of population who are illiterate or semiliterate and aged over 12 years, 1982 TOTINC81 Total income per capita, 1981 (yuan)* URBINC81 Urban income per capita, 1981 (yuan)* RURINC81 Rural income per capita, 1981 (yuan)* TOTINC79 Total income per capita, 1979 (yuan)* URBINC81 Urban income per capi:a, 1981 (yuan)* RURINC79 Rural income per capita, 1979 (yuan)* URB81 Proportion of population in urban areas, 1981. URB79 Proportion of population in urban areas, 1979 *US$1.00 = Yuan 1.71 in 1981 and US$1.00 = Yuan 1.55 in 1979.  48 N. PRESCOtT AND D. T. JAMISON availability can be measured as the ratio per 1000 population in the province. In contrast, barefoot doctors and commune health centers serve exclusively the rural population and, hence, are measured as ratios per 1000 rural population in each province. Our financial measures of health resource availability refer to budgetary appropriations through the Bureau of Public Health at provincial and lower levels (prefecture and county) of government, excluding state subsidies to the Government and Labor Insurance schemes. These budgetary outlays, measured separately for recurrent and capital expenditure per capita, are quantitatively important and account for about one quarter of estimated total health expenditure in China (Prescott and Jamison, 1984). Our measures of mortality are limited and somewhat unsatisfactory. The life expectancy data refer to levels prevailing between 1973 and 1975: they have been estimated by Young (1981) for 24 of the 29 provincial-level units on the basis of data generated by the national cancer survey, which was undertaken durinlg those years. The data on crude death rates for 1981 were generated by the Population Census carried out in 1982. More recent provincial data on life expectancy or age-specific death rates would have been preferable, but await publication of the detailed results of the 1982 Census. Our data set also includes estimates of total, urban and rural income per capita; and the percent urban population for 1979 and 1981. Since official estimates of the provincial distribution of incomes are not available, or are incomplete, we have constructed our own estimates for both 1979 and 1981. Urban income per capita is measured as the average urban wage rate in each province, weighted by the national average urban participation rate in those two years. Rural income per capita, for 1979 and 1981 respectively, is estimated by inflating provincial collectively distributed income per capita and rural house- hold expenditure per capita by the national ratios of rural net income to these variables. For each province, total income per capita is then derived as the population-weighted average of urban and rural income per capita. The data on percent urban population in each province are official estimates defined to include the population of cities, county towns, and some towns below the county level, but to exclude their agricultural sub-populations. Finally, we use the illiteracy rate in 1982 as an inverse proxy for the level of education. This measures the percent of the population who are aged over 12 years and who are illiterate or only semiliterate. The statistical procedures we use are straightforward. First, we analyze coefficients of variation in the cross-sectional data on health resource availabil- ity, in order to evaluate the distribution of these variables across provinces. In general, our hypothesis is that these variables are relatively evenly distributed given the egalitarian emphasis of China's development strategy. We then use ordinary least squares to regress these dependent variables on other indepen- dent variables in order to test hypotheses about the determinants of the interprovincial variations that we observe specifically we test and expect to reject the hypothesis that provincial differences in resource availability are systemati- cally associated with different levels of income and urbanization, and to confirm the hypothesis that mortality levels are not determined by the availability of health resource inputs (as found in some of the analyses referred to previously).  HEALTH RESOURCE AVAILABILITY IN CHINA 49 We recognize that the aggregated and cross-sectional nature of our statistical observations precludes the inference of firm causal conclusions, but the empiric- al results are nonetheless strongly suggestive. EMPIRICAL RESULTS: HEALTH RESOURCE AVAILABILITY Interprovincial variations The summary statistics, for the variables we have used, are given in Table 2. This Table shows that the range of variation, from minimum to maximum values, in our measures of provincial health resource availability is quite large. For example, the availability of total hospital beds in 1981 (HOS 81) varied from only 1.38 per 1000 in Guangxi to 4.29 per 1000 in Shanghai; while recurrent expenditure per capita in the same year (REXP 81) varied from 2.24 yuan in Anhui to 15.33 yuan in Tibet. A more revealing test of relative equality is, however, provided by examining the shape of these distributions. For this purpose, we use the coefficient of variation across provinces, measured as the ratio of the standard deviation to the mean of the distribution of provincial resource availability. The greater th:e Table 2. Provincial variables: summary statistics Mean Min Max Standard N (unweighted) deviation HOS81 2.29 1.38 4.29 0.71 29 HOS79 2.00 1.32 4.26 0.68 29 GENHOS81 1.21 0.65 2.29 0.49 29 SHW81 3.69 2.07 9.00 1.74 29 SHW79 3.30 1.79 8.28 2.55 29 DOC81 1.57 0.88 4.20 0.80 29 CHC81 0.87 0.35 1.48 0.30 29 BFD81 1.73 0.56 2.85 0.58 29 REXP81 4.39 2.24 15.33 2.84 29 CEXP81 0.77 0.27 2.41 0.56 28 LEXP75 65.53 59.25 71.97 3.65 24 CDR81 6.54 4.95 9.92 1.14 29 AUT 0.17 0 1 0.38 29 ILLIT82 36.38 15.10 78.39 14.48 29 TOTINC81 270.66 183.81 471.03 69.60 29 URBINC81 438.57 369.24 613.43 56.20 29 RURINC81 238.31 158.35 456.51 69.48 29 TOTINC79 186.34 99.63 386.22 64.15 29 URBINC79 370.08 311.58 517.63 47.42 29 RURINC79 152.59 73.51 359.40 55.22 29 URB81 0.19 0.08 0.58 0.14 29 URB79 0.18 0.07 0.57 0.14 29  50 N. PRESCOTT AND D. T. JAMISON coefficient of variation the greater the degree of inequality, and conversely. Contrary to our hypothesis these estimates, shown in Table 3, clearly indicate the existence of substantial inequalities in all types of health resource availabil- ity. The degree of inequality is less pronounced for hospital beds and health manpower, but is extreme for the distributions of both recurrent and capital expenditure per cap- Table 3. Coefficients of variation in health resource availa- bility Coefficient of variation Hospital beds: HOS81 0.31 GENHOS81 0.41 CHC81 0.35 Health manpower: SHW81 0.47 DOC81 0.51 BFD81 0.34 Health expenditure: REXP81 0.92 CEXP81 1.04 Determinants Our hypothesis of a relatively equal distribution of resources reflected the expectation that China's redistributive development strategy would have mod- ified the influence of inequalities of income level and urbanization which typically account for the unequal distribution of resource availability in other developing countries. We now provide a multivariate test of the hypothesis that these determinants are unimportant in explaining the inequality which we do observe. Our resource availability variables for each province are weighted averages of their urban and rural values which we do not observe directly. Thus, in general, our dependent variable in province i, which we denote Zi, can be written as an identity: Zi = (URBZj)URBi + RURZj (1-URBi), (1) where URBi denotes the proportion urban in the province and URBZi and RURZi denote the value of Zi in urban and rural areas, respectively. We hypothesize that the urban and rural values of Zi are linear functions of urban and rural incomes, respectively. Thus: URBZi = a()