PROGRESS TOWARD THE MILLENNIUM DEVELOPMENT GOALS IN AFRICA

PROGRESS TOWARD THE MILLENNIUM DEVELOPMENT GOALS IN AFRICA David E. Sahn Professor of Economics Cornell University [email protected] David C. S...
Author: Oscar White
16 downloads 3 Views 145KB Size
PROGRESS TOWARD THE MILLENNIUM DEVELOPMENT GOALS IN AFRICA

David E. Sahn Professor of Economics Cornell University [email protected]

David C. Stifel Research Associate Cornell University [email protected]

June, 2002

SUMMARY

We analyze Demographic and Health Surveys (DHS) to examine the progress of African countries in achieving six of the seven Millennium Development Goals (MDG) set forth by the United Nations. Our results paint a discouraging picture. Despite some noteworthy progress, the evidence suggests that, in the absence of dramatic changes in the rate of improvement in most measures of living standards, the MDG are not going to be reached for most indicators in most countries. The results are particularly sobering for rural areas, where living standards are universally lower, and where rates of progress lag behind urban areas. KEYWORDS: Africa, development goals, poverty, welfare measures, urban-rural

1

ACKNOWLEDGMENTS The authors would like to express their gratitude to the African Development Bank, the United States Agency for International Development, and the World Bank for funding this work. They would also like to thank two anonymous referees for their insightful comments, and Aparna Lhila for her skilled research assistance. Finally, they are indebted to Macro International Inc., for supplying the data, and in particular, Bridget James for her assistance and prompt responses to queries.

2

1. INTRODUCTION

The United Nations (UN) recently articulated a series of ambitious goals toward the reduction of poverty. The millennium development goals (MDG) originated from a series of UN resolutions and agreements made at world conferences held over the past decade, and were put forward with the recognition that while substantial improvements in living conditions have occurred in many countries, performances have been uneven and painfully slow in much of the developing world (UN 2000; IMF, OECD, UN and World Bank 2000). The problem of faltering social progress is especially acute in Africa, in contrast to other regions that have witnessed more sustained improvements in living standards (UNICEF 2000). Furthermore, in light of the weak and often faltering macroeconomic performances in much of sub-Saharan Africa, the prospects of continued civil conflict and vulnerability to negative shocks due to weather and related natural events, and the fact that fertility rates and population growth outpace other regions, realizing the MDG in the years ahead will be a particularly challenging task in SubSaharan Africa. Six of the seven MDG are set in clear quantitative terms. They include: (1) reducing the proportion of people living in extreme poverty by half between 1990 and 2015; (2) ensuring that all children are enrolled in primary school by 2015; (3) reducing gender inequality through eliminating the gender gap in enrollments in both primary and secondary school by 2005; (4) reducing infant and child mortality by two-thirds between 1990 and 2015; (5) reducing maternal mortality ratios by three-quarters between 1990 and 2015; and (6) ensuring that all women have access to reproductive health services by 2015. The seventh goal involves strategies to reverse loss of environmental resources through implementing sustainable development strategies. While there is some evidence on the performance of regions toward realizing these goals, the empirical evidence on how well particular countries are performing relative to these goals remains sparse. Where information does exist, it is often not based on the type of detailed empirical analysis of survey data needed to get an accurate portrayal of progress.1 In this paper, we analyze a set of reliable household survey data to examine the progress of African countries toward achieving the MDG. We are motivated by the need (a) to provide sound empirical estimates of how well African countries are doing in general, and (b) to assess the prospects of these countries toward realizing the MDG. The latter is done by extrapolating past progress, and comparing these projections to the rates of change that are necessary to realize the MDG. Throughout our analysis, we disaggregate between rural and urban areas. We find that doing so is particularly important given the substantial evidence of far worse poverty, and significantly lower living standards in rural than in urban areas in Africa. Therefore, we are particularly interested in comparing the levels and progress toward the MDG in rural areas to those in urban areas.2 A new generation of nationally representative household income and expenditure surveys has helped to provide a better understanding of living standards in Africa.

3

Prominent among these surveys is the so-called Living Standards Measurement Surveys (LSMS), which have been implemented and funded by the World Bank.3 These surveys have been enormously useful for the analysis of the level and characteristics of poverty in many African countries. In addition, there has been a series of recent studies that carefully examine changes in poverty based on household income and expenditure surveys4. While many of these studies have attempted to address issues such as variable recall periods (Scott and Amenuvegbe 1990), differences in commodity lists (Pradhan 2000), and the well-know difficulty of defining accurate deflators for inter-temporal and spatial comparisons, they are still limited in as far they (a) do not involve reliable crosscountry comparisons of well-being,5 and (b) do not address the broader issue of living standards defined over a vector of possible indicators of well-being. Thus, while we have learned a great deal about poverty, the Demographic and Health Surveys (DHS) provide us an opportunity to inform the question of how living standards across a wide-range of non-money metric indicators are evolving in Africa across a variety of dimensions. Not only have the DHS have been collected in a large number of African countries, in many cases, at more than one point in time,6 but the survey instruments are standardized for all countries, and the procedures for sampling and data collection do not vary in a substantial fashion over time. Therefore, we can confidently compare living standards across time periods, within a given country, and also across countries for many of our poverty measures. In the remainder of the paper we begin with a more detailed discussion of the methods we employ, and the variables we construct and use to evaluate progress corresponding to the six quantifiable MDG. We add a seventh goal on child nutritional status – reducing malnutrition by two-thirds – because we feel it is of great importance, and that its absence was conspicuous in the UN deliberations. In addition, we discuss the methods for making projections based on the data we have available on past performance. Section 3 then provides more details about the data, including when and where they were collected. This is followed by a discussion of the results in Section 4. We conclude with some observations about the usefulness of the goals, and our attempt to measure progress and prospects for future progress.

2. METHODS In this section we describe the indicators and methods that we use to evaluate the progress of African countries with DHS data toward achieving the MDG. We do so by addressing each goal separately. (a) The goals Goal 1: “Reduce the proportion of people living in extreme poverty by half between 1990 and 2015”

4

Given the absence of income and/or expenditure data in the DHS data, we employ factor analysis to construct an alternative measure of economic well-being – an asset index – to track poverty over time within each country.7 The assets included in this index can be placed into three categories: household durables, household characteristics and human capital. The household durables consist of indicators of ownership of radios, stereos, TVs, sewing machines, stoves, refrigerators, bicycles, and motorized transportation (motorcycle and/or cars). The household characteristics include indicator variables for sources of drinking water (piped or surface water relative to well water), toilet facilities (flush or no facilities relative to pit or latrine facilities), cooking fuel (gas or electricity), and household construction material (indicators for quality of floors). We also include the years of education of the household head to account for household’s stock of human capital. One of the properties of the asset index is that its distribution has zero mean and unit variance. And since we want to compare the distributions of assets over survey years for each country, the datasets for each of the eleven countries for which we have at least two years of survey data (and estimates of $1/day poverty rates – more on this in a minute), are pooled by country and the factor analysis asset weights are estimated for each pooled sample. They are then applied to the separate samples to estimate the asset indices for each of the households in those samples.8 To calibrate initial poverty levels, we estimate poverty lines for each of the eleven countries endogenously in order to replicate the national $/day poverty rates found in the World Development Indicators (2001).9 Because the DHS survey years and years for which we have $/day poverty estimates coincide for only Ghana and Madagascar, the poverty lines must be estimated iteratively for all of the other countries by assuming a linear rate of change in poverty between the two survey years.10 Once we have the poverty lines for each country, urban and rural poverty rates are estimated for 1990 and for the last survey year. We further project linear and log-linear poverty rate paths to the year 2015 based on “observed” changes in poverty. These are compared to linear target paths based on the stated MDG, which in the case of poverty is to cut the percent living in extreme poverty by one half. Goal 2: “Enroll all children in primary school by 2015” For ten African countries, the household roster section of the DHS data records age of individuals and their educational status for at least two survey periods.11 Using this information, we estimate the percentage of children between the ages of six and fourteen inclusive in urban and rural areas who were enrolled in school at the time of the survey. In a manner similar to the poverty estimates, we predict urban and rural enrollments in 1990 assuming a linear change in enrollment between the survey periods. Both linear and non-linear enrollment paths to the year 2015 are predicted and compared to the target path leading to the MDG of 100 percent enrollment in 2015. Goal 3: “Make progress toward gender equality and empowering women by eliminating gender disparities in primary and secondary education by 2005”

5

For the same ten countries for which we estimate changes in enrollments, we also estimate changes in the ratios of girls-to-boys enrolled in primary and secondary schools. This indicator of gender disparity in education is calculated by simply estimating in the samples of all individuals enrolled in primary and secondary schools, the ratio of girls to boys regardless of their age. These ratios (multiplied by 100) are predicted in urban and rural areas in 1990 assuming a linear change in the ratios between the survey periods. Both linear and non-linear ratio paths to the year 2005 are predicted and compared to the target path leading to the MDG of 100 in 2005. Goal 4: “Reduce infant and child mortality rates by two-thirds between 1990 and 2015” Infant mortality rates are constructed from the section of the individual survey instrument that includes birth histories of each of the women interviewed. This provides information on all live births, the ages of living children, and the dates of deaths of children who did not survive to the date of interview. Infant mortality (1q0) for a given cohort of children is defined as the simple probability of a child dying before his/her first birthday. The retrospective nature of the birth histories, however, gives rise to a censoring problem in the estimation of mortality rates. Since the birth histories are recorded for women of child-bearing age (15-49) at the time of the interview, observations on births 10 years prior to the interview do not account for children born to the cohort of women age 40-49 at that time. Sahn, Stifel and Younger (1999) find statistically significant parameters across-the-board for ten countries on the age and age squared of the mother in infant mortality regressions. Thus, uncorrected estimates of infant mortality rates become more biased as one goes back in time from the date of the survey, and are not comparable across surveys for a given time period. To avoid the censoring problem, we truncated the sample of children to only those born to mothers of age 15-39 at the date of birth, or roughly 90 percent of all children reported to have been born in each of the samples, and we extend our mortality estimates back only 10 years from the date of the survey. Infant mortality rates are estimated for cohorts of children born in each of the ten years prior to the date of the survey for the 24 African countries with DHS data.12 Note that we exclude from our sample all children born within one year of the survey because these observations represent censored spells (i.e. the child may still have died before his/her first birthday though after the enumerators visited the household).13 Regression lines are then estimated through these data points to estimate linear annual rates of change in infant mortality rates. We allow these rates of change to differ across survey years and report them as such when they are statistically different. When they do differ statistically, we use the estimated rates of change for the last survey to predict mortality rates in 1990 and in 2015. Otherwise we use the pooled estimates to predict both linear and non-linear infant mortality rate paths from 1990 to the year 2015 and compare them to the target path leading to the MDG of one-third of the mortality rate in 1990. Goal 5: “Reduce maternal mortality ratios by three-quarters between 1990 and 2015”

6

Because of the difficulty in measuring actual maternal deaths (i.e. deaths at childbirth), we employ a proxy for the prevention of such deaths. Given that a large number of maternal deaths follow from infections, blood loss and unsafe abortion, and are thus preventable, the proportion of births attended by skilled health personnel provides a means of tracking progress in preventing them. Further, since this form of health care is a primary policy mechanism that can be employed to address maternal mortality, tracking it allows us to also track the progress of public policy toward achieving this goal. Thus, while we are unable to measure the output (maternal deaths) we can and do measure changes in an input into reducing maternal mortality (births attended by skilled health personnel). This indicator of the quality of neonatal care is recorded in the maternity section of the individual survey instrument in the DHS. In this section, all women are asked about births within the five years prior to the survey, including who was present at the birth. If a doctor, a nurse, a midwife and/or a “trained health professional” was present at a birth, then the mother is recorded to have received neonatal health care from skilled health personnel for that particular birth. Since there are many mothers in the samples with more than one birth recorded in the five years prior to the surveys, it is possible (and observed) for some women to have births that were both attended and not attended by trained professionals. The percentage of births attended by skilled personnel are estimated for cohorts of children born in each of the five years prior to the date of the survey for the 24 countries with DHS data. 14 Regression lines are then estimated through these data points to estimate linear annual rates of change, and to predict the percentage of births attended by skilled health personnel in 1990 and 2015. Both linear and nonlinear paths from 1990 and 2015 are predicted and compared to the target path leading to a proxy for the MDG of reducing maternal mortality by three quarters (i.e. 90 percent of births attended by skilled health personnel). Goal 6: “Provide access for all who need reproductive health services by 2015” The DHS data have a wealth of information on knowledge and use of contraceptives. Each woman in the individual survey instrument is asked detailed questions about contraceptives as well as her current reproductive status. This permits us to estimate the share of women in need of reproductive health services who have knowledge of modern contraceptives and who use them. Two issues need clarification here. First, we define women who need access to modern contraceptives as those who are fecund and do not currently want to get pregnant. To do this, we drop from our sample of women those who are declared infecund or are menopausal, and those who report desiring to have children. This leaves non-menopausal women who either want no more children or report wanting a child but after two or more years (i.e. desiring to space the births). Second, modern contraceptives are defined as the pill, IUD, Injections, diaphragm, foam, jelly, condom, sterilization (male or female), and NorplantTM or other implants. The percentages of women in need of access to reproductive health services who (a) know of, and (b) use modern contraceptive methods are estimated for urban and rural

7

areas in the 13 African countries with at least two DHS surveys. These percentages are predicted for 1990 assuming a linear change in the ratios between the survey periods. Both linear and non-linear percentage paths to the year 2015 are predicted and compared to the target path leading to the IDF of 100 percent in 2015. Additional goal: “Reduce child malnutrition by two-thirds between 1990 and 2015” The indicator of nutritional status used in this paper (and available in the DHS) is the height-for-age of children less than 5 years of age standardized to a healthy reference population. This height-for-age z-score (HAZ) is an indicator of a child’s long-term nutritional status.15 Children who are “stunted” are those whose past chronic nutritional deprivations leave them shorter than expected for their age and gender cohorts in the reference population. In keeping with convention (WHO 1983), chronic malnutrition is defined as the percentage of the sample of children with HAZ scores below –2 (i.e. stunting rates).16 Stunting rates are estimated in urban and rural areas in the 14 countries that have at least two DHS surveys with anthropometry sections. Both linear and non-linear malnutrition paths from the last survey year for each country to the year 2015 are predicted and compared to the target path leading to a reduction in stunting rates by twothirds. (b) Determination of progress The above discussion details how we go about defining and measuring the various indicators in the MDG and sketches out how we compare progress to the goals themselves. However, in addition to comparing extrapolated progress to the goals, we address the question of whether there has been any progress in the various indicators that comprise the MDG, regardless of the question of whether the progress is consistent with the targets themselves. To do so, we conduct various statistical tests of changes over time. These are perhaps of greater importance since the goals themselves are in essence political statements, and failure to achieve the goals is not synonymous with failure to achieve social progress. In terms of the mechanics of our statistical comparisons of progress in the various indicators, we employ standard statistical tests using t-statistics for all of the goals (with the exception of infant mortality and neonatal care, where we base our tests on the statistical significance of the slope parameters estimated from our regression models). In addition, for our poverty and nutrition indicators that are derived from distributions with arbitrary cut-off points used to distinguish the poor (malnourished) from the non-poor (well-nourished), we employ standard tests of welfare dominance to compare the distributions of our nutritional status indicators and asset index over time. The idea is to make ordinal judgments on how poverty changes for a wide class of poverty measures over a range of poverty lines (Ravallion 1994 and Davidson and Duclos 2000). We discuss the concept of welfare dominance, and explain how we estimate the orderings and perform statistical inference on them in the Appendix.

8

3. DATA The Demographic and Health Survey (DHS) program has conducted over 70 nationally representative household surveys in more than 50 countries since 1984. With funding from USAID, the program is implemented by Macro International, Inc. In this study, we use 43 of the surveys for 24 Sub-Saharan African countries that have crosssectional surveys available. The DHS surveys are conducted in single rounds with two main survey instruments: a household schedule and an individual questionnaire for women of reproductive age (15-49). The household schedule collects a list of household members and basic household demographic information and is used primarily to select respondents eligible for the individual survey, though in later waves of the survey, information was also collected on educational status and attainment of all household members. The individual survey, inter alia, provides information on household assets, reproductive histories, health, and the nutritional status of the women’s young children. The quality of the data is generally good with improvements made over successive rounds. In the first wave of DHS surveys (DHS I), co-resident husbands of women successfully interviewed in the individual survey were generally also interviewed in half of the clusters. This practice was changed in the later waves (DHS II and III) to have a nationally representative sample of men, by interviewing all men aged 15-49 living in every third or fourth household. Although the designs of the surveys are not entirely uniform over time and across countries, efforts were made to standardize them, so that in most cases they are reasonably comparable.17 The DHS program is designed for typical self-weighted national samples of 5,000 to 6,000 women between the ages of 15 and 49. In some cases the sample sizes are considerably larger, and some areas are over- or under-sampled. Household sampling weights are used to account for over- and under-sampling in various regions within surveys. Since all regions are sampled in the DHS surveys, with the exception of Uganda, we make the surveys nationally representative through the use of sampling weights. Districts in northern Uganda were not included in the 1988 survey because of armed conflict. Table 1 shows the 24 African countries with DHS data and the years in which the data were collected. It also shows which indicators are available for each country. For example, all of the indicators are available for Burkina Faso, Ghana, Kenya, Madagascar, Niger, Nigeria, Tanzania, Zambia and Zimbabwe. Cameroon has all of the indicators except asset poverty because there are no estimates for $/day poverty for this country available in the 2001 World Development Indicators, and as such an absolute percentage of the population living in extreme poverty cannot be estimated using the asset index. Further, Mali has all of the indicators except those concerned with enrollments. This follows because the 1987 data was collected in the first wave in which no information was recorded on the education of the household members.18 For the nine countries with

9

only one survey, indicators are only available for changes in infant mortality and neonatal care. 4. RESULTS In this section we present the results of our analysis of progress toward the MDG in Africa. We do so by indicating whether countries are on track to realize the goals and if improvements in the various welfare measures have been observed in general during the time periods for which data are available. In other words, we compare linear target paths which illustrate the changes required per year to realize the MDG, with both linear and non-linear projections of changes over the span of time relevant to the particular MDG in question. While we have little basis for assuming that past performance is a good predictor of what will happen in the future, or that either log-linear or linear projections will reflect the evolution of the indicators that we examine, the projections are benchmarks which give a sense of where countries will be relative to the MDG if the various rates of change in the relevant indicators continue as they have to date. Before examining changes in poverty over time and how these changes fare relative to the goal of halving the percentage of those who live in extreme poverty, we first inspect the starting levels of poverty (represented by the value measured at the initial survey) in urban and rural areas. Note that in Table 2, poverty rates in each country are substantially higher in rural than urban areas, and in many cases this is difference is dramatic (e.g. Burkina Faso and Zimbabwe). This finding is consistent with findings reported elsewhere based on the use of more traditional expenditure type metrics of poverty incidence (Sahn, Dorosh and Younger 1997). The results on changes in rural poverty indicate that only in the cases of Ghana and Madagascar, do poverty rates decline at a pace that is greater than or equal to the linear trend required to realize the MDG. Having said this, for the rural areas in these two countries, if we assume diminishing gains in our projections (i.e. the log-linear projections), the target is unlikely to be reached. In other countries, particularly Mali, Nigeria and Tanzania, we find substantial declines in rural poverty, but given the high initial levels, the paces of change are not commensurate with linear projections to realize the goal. Nonetheless, as in Ghana and Madagascar, these findings of declining poverty for Mali and Nigeria are insensitive to the choice of the poverty measure or the poverty line (i.e. there is statistically significant first order dominance – see the Appendix for details).19 Kenya and Tanzania also witnessed statistically significant declines in the levels of rural poverty. However, in both these cases, the results are sensitive to the choice of poverty line and poverty measure employed (i.e. we cannot reject the null hypothesis of non-dominance). It is also worth noting that in Zambia and in Zimbabwe, statistical comparisons indicate worsening rural poverty. In the case of the Zimbabwe, we find first order dominance. Whereas in Zambia, we find statistically significant second-order improvement. In other words, when we measure poverty using the headcount ratio, we find an increase in rural poverty (at the $/day poverty line). But when we use more distributionally sensitive measures of poverty (e.g. poverty gap index or the poverty severity index) or lower poverty lines, we find that rural poverty actually

10

fell. This follows because, while the percentage of rural households in poverty rose, the well-being of the poorest households improved. In urban areas, on the other hand, poverty rates in Ghana, Kenya, Mali, Niger, Senegal and Tanzania are decreasing at paces greater than, or nearly equal to the target path. In all these countries, with the exception of Kenya with its initially very low rate of urban poverty, we also observe statistically significant declines. However, we are not able to reject the null of non-dominance in the cases of Kenya, Mali, Niger and Senegal, indicating that the results of poverty comparisons are sensitive to the choice of poverty lines and poverty measures. While we observe little if any marked change in levels of poverty in many cases, poverty worsens in urban areas in Zambia and Zimbabwe, as it did in rural areas. This is also the case in urban Madagascar.20 When we look at the data on enrollment rates for children 6 through 14 years of age (see Table 3), again we find that they are higher in urban than rural areas in each country. There is also a large divergence in enrollments across countries, with those in Niger begin the lowest, and those in Zimbabwe being the highest. In terms of changes over survey periods, we find that Kenya is the only country where the experience over the segment for which we have data puts them on a (linear and/or log-linear) path to realize the goal of 100 percent enrollments in rural and urban areas. This follows, in part, from the already relatively high enrollment rate of 77 percent in 1993. Urban Cameroon is the only other case where the linear projections are consistent with achieving the goal of universal enrollment; although this not the case if enrollment rates increase in a log linear fashion.21 The rates of improvement over the period for which we have data are also quite rapid in urban and rural Niger as well as urban Tanzania, putting them close to reaching the development goal for enrollments. Enrollments in Nigeria also increased markedly and significantly in statistical terms over the span of the nine years between the DHS surveys. We do have some worrisome findings in a few cases where the enrollment situation worsens over time. Statistically significant declines occurred in urban and rural Zambia and Zimbabwe; although, in the case of the latter, the declines are small in magnitude. Enrollment rates have also fallen in urban Madagascar over the five-year period for which data are available. In terms of the bias against girls in school enrollments (see Table 4), we find that as with other indicators, the situations during the base survey years were worse in rural areas than in urban areas in all countries, with the exception of Tanzania. Some of these urban-rural differences are large, particularly in the two Sahelian countries in our sample, Burkina Faso and Niger. For example, the ratio of girls-to-boys enrolled in rural Niger during the base survey period of 1992 is an astonishingly low 0.39. In terms of changes over time, only rural Madagascar and Tanzania are on target to meet the goal of gender equality in primary and secondary education, though this only holds for Madagascar if the observed rates of change persist to the year 2015 in a linear fashion. This is also the case in urban areas of Burkina Faso, Tanzania and Zimbabwe.

11

For Burkina Faso, this accomplishment may occur despite the initially low ratio for urban areas of 79.4, and is a result of the fact that there has been substantial progress in observed enrollments of girls in the years between the surveys. In rural Niger and Nigeria, there has been statistically significant progress as measured by increasing ratios of girls-to-boys enrolled in primary and secondary school, but the pace of change – whether we assume linear or log-linear changes – in each of these countries is below the linear target path. In a few countries, we actually observe worsening performances in terms of gender equity in enrollments. This is most pronounced in urban areas of Madagascar (where overall enrollments are also falling), Nigeria and Zambia, as well as both urban and rural Kenya. While at first glance the rates of decline in the ratios are sufficiently large to be a cause of serious concern, in fact they are only statistically significant in the case of Madagascar. Next we turn to a discussion of the results of our two health indicators, infant mortality and chronic child malnutrition, or stunting. Although the evidence for both of these indicators is mixed, the now common feature of urban areas being better off than rural areas generally applies here as well. With regard to infant mortality, we have measurements for the entire sample of 24 countries. Since we rely on retrospective recall data, we can include any country in our analysis provided that there exists at least one DHS dataset. We also can construct the longest time series, once again, owing to the reliance on recall. In terms of targets, with the exception of Kenya, and one spell in Cameroon and Zambia, all countries witnessed declines in national infant mortality rates (Table 5). When we disaggregate by region, the declining rate of infant mortality is generally greater in rural than urban areas. In the case of rural areas, however, we do not get statistically significant declines in 7 of 24 countries: Comoros, Cote d’Ivoire, Malawi, Namibia, Tanzania, Zambia and Zimbabwe. In urban areas, this applies to 16 of 24 countries, with infant mortality rising in a statistically significant fashion in Burkina Faso between 1989 and 1998, and in Zambia between 1982 and 1995. In only 11 of the 24 countries is the improvement of infant mortality in rural areas great enough to realize the goal of reducing IMR by two third by 2015 (Table 6), when linear projections are employed. Using log-linear projections, we find that in only four countries are the mortality rates falling fast enough.22 Of the 15 countries in which infant mortality rates were observed to fall in urban areas, only in Cote d’Ivoire, Ghana, Mali, and Namibia are the changes rapid enough to meet the target. In considering the results on infant mortality rates, the role and implications of the HIV/AIDS crisis clearly deserves mention. As noted above, with few notable exceptions such as urban Zambia and Zimbabwe (with particularly high rates of HIV), our empirical observations indicate that infant mortality rates have fallen during the periods for which we have data – despite the AIDS epidemic. These declines in the face of the effects of HIV/AIDS may be attributed to several factors. First, increased mortality associated with HIV had yet to contribute to large increases in overall mortality in the periods before the late 1990s. Because much of our data stops in the mid 1990s, we do not pick up many of the child deaths attributed to HIV/AIDS. Second, and perhaps even more important, the

12

overall affect of the epidemic on infant mortality is far less than on mortality rates of other age cohorts, particularly young adults. This reflects the fact that the rate of mother to child transmission of HIV in Africa is estimated to be approximately 30 percent (Preble, 1990). Further, among those cases where there is positive transmission, the probability of dying during the first year of life from AIDS is estimated to be only 15 percent (Preble, 1990). So consider, for example, a country where the prevalence rate among pregnant women is high (15 percent). The impact of mother-to-child transmission of HIV on infant mortality will be less than 7 deaths per 1000 live births. If overall infant mortality is 100 or above, as it is in many African countries, the impact of HIV/AIDS as a percent of total infant mortality will be limited relative to the deaths from other causes. As we look ahead, the impact of HIV/AIDS on infant mortality is also quite uncertain. On the one hand, we recognize that – given the ravages of HIV/AIDS – making projections based on the changes in mortality during the late 1970s through the early 1990s may not be meaningful in much of sub-Saharan Africa. On the other hand, in the medium to short-term, there are reasons to be hopeful that anti-retrovirals (ARVs) will substantially reduce mother-to-child transmission. This should reduce the number of AIDS-related deaths among infants and contribute to acceleration in the improvements witnessed during the past few years, or at least should not reverse the trend. In addition, there is some reason to hope that behavioral changes will lead to a reduction in the share of pregnant women with HIV. But of course, important obstacles remain in terms of making widely available the ARVs and other interventions that may lower adult prevalence rates. In any event, in Table 6 we have shown an alternative HIV-adjusted predictive path for infant mortality that is based on four strong assumptions. First, we assume that HIV prevalence rates among pregnant women remain fixed at the most recently estimated levels. (The Appendix Table presents the seroprevalence rates for pregnant women in urban and rural areas.) Second, we assume that there is no reduction in mother to child transmission rates. This assumption, coupled with evidence that 15% of the HIV-positive children will die in their first year of life leads to increased levels of mortality associated with HIV/AIDS (also shown in the Appendix Table). Third, we assume that none of the children who die of HIV/AIDS would have died due to other causes. Finally, we assume that for the period in which we have data, there are no HIV related deaths (since we are assuming that the AIDS shock is additional to the changes that have already occurred in the IMRs). This latter assumption is clearly the most extreme. Further, it is most obviously not correct in such cases as Zimbabwe and Zambia where the epidemic hit early and hard, and where the increased IMRs undoubtedly already capture the effect of mother-to-child transmission. The results indicated that the overall impact of our HIV adjustment on the IMR projections is quite modest in most cases. Obvious exceptions include those countries with high seroprevalence, such as Namibia and Malawi. In the case of the former, the log-linear projection for IMR in urban areas is 31 using our HIV adjustment, versus only 21 without. And in urban Malawi, the comparable figures are 181 based on the assumptions above, versus only 169 otherwise.

13

Our examination of changes in health outcomes using child anthropometric measures provides some additional insights that are not always consistent with the story we get from looking at changes and levels of infant mortality. We employ the same target as is used for mortality: a two-thirds reduction of the percent of children who are stunted. In Table 7, we find that the percent of malnourished children is greater in rural areas than urban areas in all countries for which we have data. Only in urban areas in Ghana, Tanzania and Togo, as well as rural areas in Togo, do we observe declines in stunting that are consistent with meeting the target (both linearly and log-linearly). However, the in case of Tanzania, we are unable to reject the null of non-dominance when comparing the levels of malnutrition between the two periods, and we are only able to do so using second order conditions in urban Ghana and rural Togo, despite the fact that the point estimates are all statistically different. In rural areas of Uganda and Zimbabwe, the rates of malnutrition fall, but not fast enough to keep pace with linear projections needed to cut malnutrition by two-thirds. Nevertheless, in both cases, we are able to reject the null of non-dominance implying unambiguous improvements in the nutrition of the pre-school age child populations in these countries. In many countries, malnutrition shows no improvement or actually worsens. Malnutrition in rural and urban areas in Burkina Faso, Cameroon, Mali, Niger, and Nigeria, as well as rural Madagascar, Senegal and Zambia deteriorated between over the periods spanned by the respective surveys. We now turn to the analysis of two MDG for which there are no precise measures in the DHS: access to reproductive health services and maternal mortality. In the case of maternal mortality, we have a reasonable, but far from perfect proxy measure – the number of births attended by skilled health personnel. Like infant mortality, these data come from recall information that allows us to construct a longer data series than the periods between surveys. Before discussing the prognoses for realizing the development goal, we should first highlight the large disparities in the percent of births attended by qualified personnel both across countries and between rural and urban areas (Table 8). The figures from rural Chad and Niger paint a particularly acute picture. In the last years for which have survey data, only 4.9 percent and 7.8 percent of births in rural areas of these countries, respectively, were attended by qualified medical personnel. Contrast this with rural Zimbabwe where the comparable figure was over 62.6 percent. In urban areas in Burkina Faso and Zimbabwe, qualified persons attend over 90 percent of the births, and in a number of other countries the figures approach this level. We should also add that the high value for Burkina Faso is surprising. We admonish that determining whether personnel are “qualified” or “skilled” is a subjective undertaking, and consequently we should treat cross-country comparisons of this indicator with caution, despite the comparability in the questionnaire and training of enumerators in the various surveys. In examining and interpreting our results relative to the subject of interest – maternal mortality – we should bear in mind that progress in increasing the number of qualified birth attendants should be easier to achieve than lowering maternal mortality

14

rates. Thus, it is not a good sign that the only cases where we find statistically significant improvements consistent with the linear rates of progress required to realize the targets are urban areas in Cameroon, Mali, Mozambique, Senegal and Togo. There is virtually no realistic hope for rural areas in any of the countries to meet the goal. Although, rural areas of Comoros, Ghana, Mali, Namibia, Niger and Senegal, all witnessed statistically significant improvements in this indicator, the low starting levels and slow rates of improvement will keep them from realizing the goal regardless of our assumption about curvature in the rates of improvement. In Tables 9 and 10, we present two proxies for access to reproductive health services. Given the difficulty of disentangling access from knowledge and usage, we present estimates of the percent of women who (a) know of, and (b) use modern forms of contraceptives. As described in the methodology section, these percentages are calculated over the population of women who “need” access (i.e. fecund, nonmenopausal and not desiring to currently have children), not over all women. Our findings show that in both urban and rural areas in every African country for which we have data (except Madagascar, which nevertheless has seen statistically significant improvements in rural and national levels), knowledge of modern contraceptives has been increasing at linear rates in excess of the target path that would results in the goal of 100 percent access being realized. The rapid increases in rural Mali, and to a lesser extent rural Nigeria, are particularly pronounced. In contrast, there is no case in either urban or rural areas, where the percent of women using modern methods of contraception increases at a rate that even comes close to following the target path (Table 10). Aside from Kenya and Zimbabwe, in none of the African countries for which we have data do we find more than 10 percent of rural women using modern contraceptive devices in the first survey period. The picture is not much rosier in urban areas. While nearly two-thirds of the women in urban Zimbabwe use such contraception, the next highest figure is 35 percent from urban Kenya. In no other country did more than one-quarter of the urban women use modern contraceptives in the first survey period. Thus, while this is consistent with the general picture of higher living standards in urban relative to rural areas that we observe for other indicators, use of modern contraceptives remains low in rural and urban areas alike. Despite this sobering assessment of the bleak prospects for realizing the goal of access to modern contraception, we find statistically significant improvements in all cases except urban Madagascar and urban Zimbabwe. The fact that these improvements have occurred despite the poor performance relative to targets reflects the ambitious nature of the target to ensure universal access, especially given the low starting levels. This situation contrasts with the previous indicator of knowledge of modern contraception. For example, in Burkina Faso, only 2.4 percent of the women in rural areas reported using modern contraception at the base period for which we have data (1992), while just over 60 percent indicated they were knowledgeable of modern contraceptive methods. This is an enormous difference that may reflect largely the inaccessibility of services, but may be a result of other socio-cultural factors as well. From the perspective of our target analysis, the slope of the target path is thus far steeper for every country when it comes to

15

realizing universal use of modern contraception than universal knowledge. In other words, far greater absolute changes in use than knowledge are necessary to meet these goals. This being said, we find in Tables 9 and 10 that, except for Kenya, the absolute increases in the percent of women with knowledge of modern contraception exceeds the absolute increases in the percent of women using modern methods, despite the base level for the latter being far lower. In urban areas, this is also the case for Cameroon, Ghana, Mali and Nigeria. In Madagascar, both use and knowledge fell as a percent of urban women in need of reproductive health services. In those urban areas where the absolute percentage increase in knowledge of modern contraceptives was less than the increase in usage, the initial percentages of women with knowledge of modern methods were 90 percent or higher, thus there is little room for improvement. In Tables 11 we summarize the results of our target analysis, by country for rural, urban and national levels. The first set of columns show, when using linear projections, the number of indicators that are on target to achieve the development goal as stated in the MDG (and the added goal of reducing malnutrition by two-thirds) for each country. Note that we have data on five or more goals for 12 countries.23 Among these countries, in rural areas for Burkina Faso, Mali, and Niger, changes already observed in the data suggest that none of the MDG will be met. To make matters worse, in Nigeria and Zambia, this also holds true in urban areas. In rural areas, Ghana and Madagascar are the only two countries to be on target to achieve two of the seven goals. Admittedly, we have information on the realization of only two goals in nine countries. Nonetheless, it is only in rural areas in these latter countries where we find that at most half of the goals are likely to be met. In some cases, though not all, urban areas are faring better than rural areas in terms of their potential for reaching the targets.24 For example, in urban Ghana and Tanzania, three of the seven targets are likely to be achieved, with two of them being reductions in poverty and malnutrition. In Senegal and Mali, where we only have measurements on five of the seven goals, three of the targets are likely to be reached in the urban areas (again with a commonality of declines in poverty). Nonetheless, in the majority of the cases, the MDG are unlikely to be met in urban, just as in rural, areas in Africa. A somewhat more optimistic assessment of progress toward realizing improvements in living standards is found in the last three columns of Table 11, which simply address whether there are statistically significant improvements for the indicators, regardless of whether the rates of improvement are rapid enough to reach the MDG. At a national level, Ghana, Madagascar and Niger have witnessed improvements in 5 out of 7 indicators. A number of countries, particularly Senegal, Togo, and Uganda also do well in terms of the share of indicators for which improvements are noted. In contrast, Burkina Faso, Zambia and Zimbabwe only see improvements in 1 out of 7 goals. In rural and urban areas, Ghana and Niger are also the best performers of those countries where we have data on all the goals, but again, Senegal, Togo and Uganda do well in both urban and rural areas, as does Mali in urban areas.

16

5. CONCLUSIONS In this paper we have quantitatively examined progress toward the MDG among African countries, in addition to making statistical comparisons of changes in living standards over time. The results paint a discouraging picture. Despite some noteworthy progress in improved living standards in certain dimensions in a select number of countries, the preponderance of the evidence suggests that, in the absence of dramatic changes, the MDG are not going to be reached for most indicators in most countries. In the case of our poverty estimates, only two of 11 rural populations exhibit progress commensurate with the linear target path; although, in five countries, the urban sectors are on target. With regard to enrollments, only Kenya and urban Cameroon are on target. Things would even be worse if we assume that the paths of improvement over time take on log-linear shapes. Three of the ten countries for which we have data on changes in education have witnessed reductions in gender discrimination in urban areas consistent with the target. What is worse is that this is only the case for rural areas in two countries. Not a single country out of 24 is on target in terms of increasing births attended by skilled personnel in rural areas, and only four are on target in urban areas. The HIV/AIDS epidemic, in fact, will likely make attended births all the more important, especially to the extent that mid-wives and other health care workers have access to pharmaceuticals to reduce the probability of mother to child transmission. Likewise, other harmful consequences of the failure to progress in this area include jeopardizing the health of women during delivery, increasing the chances of neonatal deaths, and losing the opportunity for birth attendants to educate women in areas such as proper child lactation and weaning practices, and child spacing technologies. Togo is the only country of the 14 for which we have anthropometric data over time, where the target for reducing rural malnutrition seems realizable. Things are slightly better in urban areas, where Ghana and Tanzania are also on target. These stunting results are robust to our assumptions about whether rates of change assume a linear or log-linear path through 2015. In terms of universal access to modern reproductive health services, not one country is on track vis-à-vis use of modern contraceptives. In most countries, however, knowledge of modern contraceptive methods is likely to be universal by 2015 provided that the trends continue linearly. If, however, we assume that the trajectories are log-linear, things do not look nearly so favorable in rural areas. Fortunately, all this bad news is slightly tempered by the trends in infant mortality. The results of our analysis suggest that in ten (six) of 24 countries, mortality targets will be realized in rural (urban) areas if improvements persist linearly. More encouraging is our analysis of improvements in living standards in general, regardless of whether the progress is sufficient to realize the MDG. For many indicators, we observe statistically significant improvements in many, if not all the countries, even when the current rates of change will leave the targets unmet by 2015. For example, no country is on target to realize the goals regarding modern contraception, but all have made progress over the periods for which we have data. Most countries have reduced

17

poverty, but few are on target to halve poverty by 2015. More than half the countries have increased enrollments, but only one is on target to reach the goal of universal enrollment. Finally, although only one country is on target to reduce malnutrition by two-thirds, four of 14 countries have reduced levels of stunting in the 1990s. So, while this is welcome news, legitimate concerns remains about the pace of progress. Quite simply, statistically significant improvement is just that, and says little about the rate of change. While our findings are particularly informative with regard to assessing the progress of countries toward achieving the MDG, it is important to keep in mind that in most cases, we rely on only two survey time periods from which we can estimate trends. Further, these cover only a relatively brief span of time. As more DHS surveys come on line, it will be important to update our analysis in order to extend the period of coverage. Nonetheless, despite the major pitfall of our work – that the time periods for which we can measure change are often brief (with the exception of the mortality statistics) – we argue that the information on inter-temporal trends based on only two DHS surveys is far more accurate and meaningful than similar two point in time comparisons for, say, changes in income and/or expenditures. This follows because, first, survey and sampling methods do not change for the DHS surveys we employ. Second, we need not worry about problems of deflators, purchasing power parity, and other market price related issues that plague money-metric welfare measures. Third, for some indicators such as infant mortality and attended births, we can construct long-term data series from retrospective information. Fourth, for most other indicators, there are unlikely to be significant transient or short-term intertemporal fluctuations. For example, contraceptive knowledge and chronic malnutrition (the cumulative effect of well-being over the past few years) will not change markedly in response to the types of shocks that may affect incomes. Thus, our trends are based not on two unstable snapshots of well-being, but rather on measures the stock of well-being at the time of the surveys. We believe that these stocks are not terribly sensitive to short-term exogenous shocks. This being said, we still need to emphasize that for some indicators, particularly mortality, the notion of linear progress, and its pace, are conditioned by the ravages of HIV/AIDS, and other unforeseen positive and negative health shocks. Thus, while we compare log-linear and linear projections to goals, the true shape of path of progress is unknown. Are there increasing returns to investing in health and education infrastructure? Is it reasonable to expect the pace in declining mortality to accelerate or decelerate in most countries? Will diseases like tuberculosis continue their resurgence (WHO, 2001)? Will the prevalence of other diseases like malaria continue to fluctuate (WHO, 1999)? We do not claim to have satisfactory answers to these questions. Ultimately, regardless of the assumptions we make in our analysis, the poor performance in terms of improving living standards in Africa seems irrefutable. In fact, if we consider that the worst-off countries in sub-Saharan Africa (i.e. those torn apart by war and communal violence) are not represented in the survey data analysis, things overall are undoubtedly worse than portrayed here. The results are particularly sobering for rural areas, where living standards are universally lower, and where rates of progress generally lag behind urban areas. In the final analysis, the targets against which we

18

compare progress are the outcome of political consultations among international organizations and member countries of the United Nations. Even if they were adjusted to be somewhat less ambitious, the basic results of our analysis would not change. Rather, given the rate of progress, of which there has been some, quite substantial revisions in the MDG will likely be necessary to avoid widespread failure to achieve targets across the continent. This being said, we do not address perhaps the fundamental issue: how policy can be re-shaped to accelerate progress toward the MDG. The paramount importance of sustainable pro-poor growth, and the strategic pillars necessary to achieve it have been widely debated and articulated. As such, an assessment of these policies is beyond the scope of this paper. Nonetheless, because the targets are out there and are widely disseminated and used by governments and the international agencies, an assessment of progress toward them and expectations about them are extremely important. This paper is one such reality check, with results that are not terribly encouraging for those concerned about raising living standards in Africa.

19

References Alwang, J. (2000). Structural adjustment, drought, and poverty in Zimbabwe: Changes in 1990s. Mimeo, Poverty Reduction and Social Development, Africa Region, The World Bank, Washington, D.C. Alwang, J., & Ersado, L. (1999). Changes in poverty in Zimbabwe: 1990-1996. Mimeo, Poverty Reduction and Social Development, Africa Region, The World Bank., Washington, D.C. Alwang, J., Mills, B., & Taruvinga, N. (1999). Changes in well-being in Zimbabwe, 1990-1996: Non-parametric evidence. Mimeo, Poverty Reduction and Social Development, Africa Region, The World Bank, Washington, D.C. Appleton, S., Emwanu, T., Kagugube, J., & Muwonge, J. (1999). Changes in poverty in Uganda, 1992-1997. Mimeo, Poverty Reduction and Social Development, Africa Region, The World Bank, Washington, D.C. Burger, K., Hoogeveen, H., Kinsey, B., & Sparrow, R. (2000). Welfare and poverty among rural households in Zimbabwe. Mimeo, Poverty Reduction and Social Development, Africa Region, The World Bank, Washington, D.C. Center for International Research, U.S. Bureau of the Census. (2001). HIV/AIDS Surveillance Data Base. Washington, D.C.: U.S. Bureau of the Census. Coulombe, H., & McKay, A. (2001). The evolution of poverty and inequality in Ghana over the 1990s: A study based on the Ghana Living Standards Surveys.” Mimeo, Office of the Chief Economist, Africa Region, The World Bank, Washington, D.C. Davidson, R., & Duclos, J-Y. (2000). Statistical inference for stochastic dominance and for the measurement of poverty and inequality.” Econometrica, 68, 1435-1465. Dercon, S. (2000). Changes in poverty and social indicators in Ethiopia in the 1990s: (At last) Some good news from Ethiopia.” Mimeo, Poverty Reduction and Social Development, Africa Region, The World Bank, Washington, D.C. Dercon, S. (2001). The impact of economic reforms on households in rural Ethiopia 1989-1995. Mimeo, Office of the Chief Economist, Africa Region, The World Bank, Washington, D.C. Eastwood, R., & Lipton, M. (2000). Rural-urban dimensions of inequality change. World Institute for Development Economics Research (WIDER) Working Papers No. 2003, Helsinki, Finland.

20

Filmer, D., & Pritchett, L. (2001). Estimating wealth effects without expenditure data — or tears: An application of educational enrollment in states of India.” Demography, 38, 115-132. Foster, J., & Shorrocks, A. (1988). Poverty orderings. Econometrica 56, 173-177. Hammer, J. (1998). Health outcomes across wealth groups in Brazil and India. Mimeo, DECRG, The World Bank, Washington, D.C. International Monetary Fund, Organisation for Economic Co-operation and Development, United Nations, &World Bank. (2000). A better world for all: progress towards the international development goals. A joint report that can be downloaded from the web: http://www.paris21.org/betterworld. McCulloch, N, Baulch, B., & Cherel-Robson, M. (2000a). Poverty, inequality and growth in Zambia during the 1990s. Mimeo, Poverty Reduction and Social Development, Africa Region, The World Bank, Washington, DC. McCulloch, N, Baulch, B., & Cherel-Robson, M. (2000b). Growth, inequality and poverty in Mauritania, 1987-1996. Mimeo, Poverty Reduction and Social Development, Africa Region, The World Bank, Washington, D.C. Preble, E. (1990). Impact of HIV/AIDS on African children. Social Science Medicine, 31, 671-680. Pradhan, M. (2000). How many questions should be in a consumption questionnaire? Evidence from a repeated experiment in Indonesia. Cornell Food and Nutrition Program Working Paper No. 112, Cornell University, Ithaca, NY. Ravallion, M. (1996). Poverty comparisons. Chur, Switzerland: Harwood Press. Razafindravonona, J., Stifel, D., & Paternostro, S. (2001). Changes in poverty in Madagascar: 1993-1999. World Bank Africa Region Working Paper Series, No. 19, Washington, D.C. Sahn, D. (Ed.) (1989). Seasonal variability in third world agriculture: The consequences for food security. Baltimore: Johns Hopkins University Press. Sahn, D., Dorosh, P., &Younger, S. (1997). Structural adjustment reconsidered: Economic policy and poverty in Africa. Cambridge, UK: Cambridge University Press. Sahn, D., & Stifel, D. (2000). Poverty comparisons over time and across countries in Africa. World Development, 28, 2123-2155.

21

Sahn, D., & Stifel, D. (2001). Exploring alternative measures of welfare in the absence of expenditure data. Cornell Food and Nutrition Policy Program Working Paper, No. 97, Cornell University, Ithaca, NY. Sahn, D., David Stifel, and Stephen Younger. 1999. “Inter-Temporal Changes in Welfare: Preliminary Evidence from Nine African Countries.” Cornell Food and Nutrition Policy Program Working Paper, No. 94. Ithaca, NY. Sahn, David, Stifel, D., & Younger, S. (2002). Evidence on the urban-rural divide in Africa: Should we be concerned?” A paper prepared for the Oxford University Conference on “Spatial Inequality in Africa”, September 21-22 ,2002. Scott, C., & Amenuvegbe, B. (1990). Effect of recall duration on reporting of household expenditures: An experimental study in Ghana. Social Dimensions of Adjustment in Sub-Saharan Africa Working Paper 6, World Bank, Washington, D.C. Strauss, J., & Thomas, D. (1995). Human resources: empirical modeling of household and family decisions. In Handbook of development economics vol. 3A (pp.18832023). New York: Elsevier Science, North Holland. United Nations. (2000). Millennium summit goals. New York: United Nations. UNICEF (2000). The state of the world’s children 2000. Available on-line at http://www.unicef.org/sowc00/ World Bank. (2001). 2001 African development indicators. Washington, D.C.: World Bank. World Bank. (2001). 2001 World development indicators. Washington, D.C.: World Bank. World Health Organization (WHO). (1983). Measuring change in nutritional status: guidelines for assessing the nutritional impact of supplementary feeding programmes for vulnerable groups. Geneva: WHO. World Health Organization (WHO). (1995). Physical status: The use and interpretation of anthropometry. A report of the WHO expert committee. Geneva: WHO. World Health Organization (WHO). (1999). Malaria, 1982-1997. Weekly Epidemiological Record., No. 32. World Health Organization (WHO). (2001). WHO report 2001. Geneva: WHO.

22

APPENDIX Consider two distributions of welfare indicators with cumulative distribution functions, FA and FB , with support in the nonnegative real numbers.25 Let D 1A ( x) = FA ( x) = ∫ dFA ( y ) . x

0

If D 1A ( x) ≤ (

Suggest Documents