QUATRE ESSAIS SUR LES LIENS ENTRE LA PAUVRETE, L'INEGALITE ET LA SANTE AVEC UNE APPLICATION EMPIRIQUE AUX PAYS EN DEVELOPPEMENT: L'AFRIQUE COMPAREE AU RESTE DU MONDE

In this dissertation, we are mainly interested in the interactions between poverty and one of its greatest dimensions1, namely health. More specifically, we will focus on their inequalities: does poverty inequality have more effect on poverty than health level? Does health inequality matter to poverty? Poverty and health are two related concepts that both express human deprivation. Health is said to be one of the most important dimensions of poverty and vice-versa. That is, poverty implies poor health because of a low investment in health, a bad environment and sanitation and other living conditions due to poverty, a poor nutrition (thus a greater risk of illness), a limited access to, and use of, health care, a lower health education and investment in health, etc2. Conversely, poor health leads inevitably to poverty due to high opportunity costs occasioned by ill-health such as unemployment or limited employability (thus a loss of income and revenues), a lower productivity (due to loss of strength, skills and ability), a loss of motivation and energy (which lengthen the duration of job search), high health care expenditures (or catastrophic expenditures), etc3. But what are the degree of correlation and the direction of the causality between these two phenomena? Which causes which? This is a classic problem of simultaneity that has become a great challenge for economists. Worst, each of these phenomena (health and poverty) has many dimensions4. How to reconcile two multidimensional and simultaneous events? 1 Aside the income-related material deprivation. 2 Tenants of the ?Absolute Income? hypothesis for instance show that absolute income level of individual has positive impact on their health status (Preston, 1975; Deaton, 2003). Conversely, lack of income (and the poverty state it implies) leads unambiguously to poor health. For other authors, it is not the absolute level per se, but the relative level (i.e. comparably to others in the society) that impacts most health outcomes. This is the ?Relative Income? hypothesis (see van Doorslaer and Wagstaff, 2000, for a summary). 3 See Sen (1999) and more recently Marmot (2001) for more information. 4 Poverty could be seen as monetary poverty, human poverty, social poverty, etc. Identically, one talks of mental health, physical health, ?positive? and ?negative? health, etc. So a one-on-one causality could not possibly exits between the two, or will be hard to establish. We?ve chosen the first way of causality: that is, poverty (and inequality) causes poor health. As justification, we consider a life-cycle theory approach (Becker, 1962). An individual is born with a given stock of health. This stock is supposed to be adequate enough. During his life, this stock is submitted to depreciation due to health shocks and aging (Becker?s theory, 1962). We could think that the poorer you are, the more difficult is your capacity to invest in your health5. Empirically, many surveys (too numerous to be enumerated here) show that poor people6 do have worse health status (that is, high mortality and morbidity rates, poor access to health services, etc.). It has been established that poor children are less healthy worldwide, independently of the region or country considered. It is generally agreed that the best way to improve the health of the poor is through pro-poor growth policies and redistribution. Empirically, one of the major achievements of these last two decades in developing countries is the improvement in health status of populations (notably the drop in mortality rates and higher life expectations) following periods of (sustained) economic growth. However, is this relation always true? In some countries as we will see later in this thesis, while observing an improvement in the population?s welfare, the converse is observed in its health status, or vice versa. If health and poverty are so closely related, they should theoretically move in the same direction. But the fact that in some countries we observe opposite trends suggests that some dimensions of health and poverty are not or may not be indeed so closely related, as postulated, and that they may depend of other factors. 1. The Purpose of the Study. 5 Another justification is that many authors have studied the problem the other way. Schultz and Tansel (1992, 1997) for instance showed that ill-health causes a loss of revenues in rural Cote d?Ivoire. Audibert, Mathonnat et al. (2003) also showed that malaria caused a loss of earnings of rural cotton producers in Cote d?Ivoire. 6 Usually defined from some income or expenditure-related metric or some assets-based metric. The ultimate goal of our dissertation in its essence is to measure inequality in health7 in developing countries using mainly Demographic and Health Surveys (DHS, henceforth)8. It deals with interactions between poverty and one of its greatest dimensions, putting aside the income-related material deprivation, namely health. It therefore measures inequality in health status and access to health and discusses which policies should be implemented to correct these inequalities. That is, it aims to see how much rich people are better off and benefit from health interventions, as compared to the poor, and how to reduce such an inequality. The present dissertation contains four papers that are related to these questions. Our main hypothesis (that will be tested) is that poverty impacts health through inequality effects9. Graphically, we can lay these simple relationships as: The dashed line in the figure above suggests that income inequality could impact health directly. But we consider that this direct effect is rather small or negligible, as compared to the indirect effect through inequality in health. Therefore, inequality in health is central to our discussion. To measure inequalities in health, we face three challenges: 7 And corollary health sanitation (access to safe water, toilet and electricity). Though electricity is more a measure of economic development that health measure per se, we add it here as a control for sanitation and nutrition: for example women could read more carefully the drugs? notices, or warm more quickly foods; more generally, electricity often improves the mental and material wellbeing of households. It also conditions health facility?s performance. 8 And potentially other surveys. In this case, we mention explicitly the survey(s). 9 The other important factor that could impact health is the performance of the health system. This is discussed in the Chapter 3. Health Assets Inequality Health Inequality Poverty (Assets Index) - measuring welfare (income metric) and subsequently inequality in welfare, - measuring health, - and measuring inequality in health. The measurements can be conducted using two approaches (Sahn, 2003): - Directly by ranking the households or individuals vis-à-vis their performance in the health indicator; we thus have a direct measure of inequality in health. This is suitable when the health indicator is continuous (such as weight, height or body mass index). According to Prof. David E. Sahn, that approach ?which has been referred to as the univariate approach to measuring pure health inequality, involves making comparisons of cardinal or scalar indicators of health inequality and distributions of health, regardless of whether health is correlated with welfare measured along other dimensions?. - Indirectly by finding a scaling measure such as consumption or income or another indicator (assets index for instance)10 that would help ranking the households or individuals (from the poorest to the richest), and see what are their performance in the health variable of interest. We are therefore measuring an indirect health inequality. The indirect method is mostly suitable when the health indicator is dichotomous (for example whether the individual has got diarrhoea last 2 weeks, or ?have the child been vaccinated?, or ?place of delivery?) or is a rate (such as child mortality). Again, quoting Prof. Sahn, ?making comparisons of health across populations with different social and economic characteristics is often referred to in the literature as following the so-called `gradient? or `socioeconomic? approach to health inequality. Much of the motivation for this work on the gradient approach to health inequality arises out of fundamental concerns over social and economic justice. The roots of the gradient will often arise from various types of discrimination, prejudice, and other legal, social, and economic norms that may contribute to stratification and fragmentation, and subsequently inequality in access to material resources and various correlated welfare outcomes?. While the first method would appear quickly limited for dummy or limited categorical health variables because of the small variability in the population, the second approach could also be 10 Or more generally any other socioeconomic gradient such as education, gender or location. impossible when no information is available to scale the units of observation in terms of welfare. We?ll be mostly focusing on the second approach, as did many health economists, and also due to the nature of the DHS datasets in hand and the indicators that we are investigating. 2. Strategy, Methods and Structure. Measuring wealth-related inequality in health implies in the first stage defining and characterizing the poor. Who are indeed the poor? Traditionally, monetary measures (income or consumption) have been used to distinguish households or people into ?rich? and ?poor? classes. Indeed, it is agreed that the ?incomemetric? approach is one of the best ways to measure welfare11. However, it sometimes, if not often, happens that we lack this essential information in household survey datasets. Especially in our case, the DHS datasets do not have income nor consumption information. Then, how to characterize the poor in this situation? For a long time, economists have eluded the question. But soon, it became evident that an alternative measure is needed to strengthen the ?poverty debate?. In the first part of our dissertation, we start by providing a theoretical framework to find a proxy for wellbeing, in the case where consumption or income-related data are missing, namely by discussing the use of assets as such a proxy. The first part of this thesis is relatively long, as compared to the second. However, this is justified, due to its purpose. The goal of the first part of the dissertation is to participate to the research agenda on poverty. It attempts to measure it in a ?non traditional?12 way. 11 There is a consensus in the economic literature that income is more suitable to measure wealth or welfare in developed countries while consumption is more adequate for developing ones due to various reasons such as irregularity of incomes for informal sector, seasonality, prices, recall periods, trustworthy, etc. (see Deaton 1998 for detail). 12 i.e. a non monetary way. The main rationale for this first part therefore is thus to find a new, non monetary measure to characterize in best, life conditions, welfare and then the poor. This measure is referred to as the ?assets index?. Indeed, as the majority of developing countries are engaged more and more in fighting poverty, inequality and deprivation, more and more information on the state of poverty13 is needed. If in almost all these countries, many household surveys have been implemented to collect information on socioeconomic indicators, the major indicator that is needed to analyze poverty (namely income or consumption data) is unfortunately not often collected due to various reasons (time, cost, periodicity, etc.). Even, if they were collected, the quality of the data is often poor. Therefore, economists tend to rely more on other indicators to compensate for the absence of monetary measures. One of the indicators often used are the assets owned by households. The question arose then how to use these assets to characterize the poor in this context? How to weight each of them? In a first attempt, many economists built a simple linear index by assigning arbitrary weights to the assets14. In a seminal paper, Filmer and Pritchett (2001) propose to construct the so-called ?assets index? which could be used as a proxy for consumption or income. It is commonly agreed that their methodology follows a ?scientific? approach, thus is more credible. In their case, they use a Principal Component Analysis (PCA, henceforth) to build their assets index. Since, many other economists have followed in their footsteps which we label in our dissertation, the ?material? poverty approach (as opposed to the monetary one) since it is based on materials (goods and assets) owned by the households or individuals. Because of the importance of the subject (poverty) and because the method is pretty new and original, this first part of our thesis is as said quite long as compared to the second one and has two papers which focus mainly on poverty and inequality issues and their connections with economic growth. In this part, we start by presenting a methodology of measuring non monetary (material) poverty, when a consumption or income data is not available. We show how one can obtain robust results using assets or wealth variables. Once the method is clearly 13 And more generally welfare. 14 For example a television is given a weight of 100, a radio 50, and so on. But this is clearly not a `scientific? way to proceed, as there is no rational ground in giving such weights. tested and validated, it is then confronted to real data. We show that the index shares basically the same properties with monetary metrics and roughly scales households in the same way as does the consumption or income variables. We discuss the advantages and also the limitations of using the assets index. The important thing to bear in mind is that, once it is obtained, it could be used to rank the observational units by wealth or welfare level. - The first chapter defines in a first section poverty and how it is usually measured (by the income metric approach). We discuss the limitations of the use of income/expenditure and what could be alternative measures. We then propose in section 2 the assets metric as a proxy for poverty measurement and test the material poverty approach on international datasets collected by the DHS program. We explore the material poverty and inequality nexus in the world and how Sub-Saharan Africa (SSA)15 compares with other regions. We show, using that index and DHS data, that poverty, at least from an assets point of view, was also decreasing in SSA as well as in other regions of the world. This result contrasts with other findings such as Ravallion and Chen (2001) or Sala-i-Martin (2002) that show that, while other regions of the world are experiencing a decline in their (monetary) poverty rates, SSA is lagging behind, with rates starting to rise over the last decade. Therefore, two different measures of welfare could yield opposite results and messages in terms of policies to implement to combat poverty. Moreover, the method we use not only allows observing changes over time for each country, but also provides a natural ranking among countries (from the poorest to the richest). In this chapter, aside the measure of welfare and poverty, we also discuss in a final section the impact of demographic transition on economic growth and therefore on poverty. Indeed, demographic transition is a new phenomenon that is occurring in developing countries, especially African ones. Its negligence could lead to underestimating poverty measures (both material and monetary) by underestimating real economic growth rates. We show that changes in the composition and the size of households put an extra-pressure on the development process. While traditional authors have not considered the impact of these 15 SSA countries are Benin, Burkina Faso, Central African Republic, Cameroon, Chad, Comoros, Republic of Congo, Côte d?Ivoire, Ethiopia, Gabon, Ghana, Guinea, Kenya, Lesotho, Madagascar, Malawi, Mali, Mauritania, Mozambique, Namibia, Niger, Nigeria, Rwanda, Senegal, South Africa, Tanzania, Togo, Uganda, Zambia and Zimbabwe. The ?rest of the world? is represented by Armenia, Bangladesh, Bolivia, Brazil, Colombia, Dominican Republic, Egypt, Guatemala, Haiti, Honduras, India, Indonesia, Kazakhstan, Kyrgyz Republic, Moldova, Morocco, Nepal, Nicaragua, Pakistan, Paraguay, Peru, Philippines, Turkey, Uzbekistan, Vietnam and Yemen. changes, we show that taking this into account implies higher economic growth rates than those actually observed or forecasted. - Once the assets index approach is established and tested on international data, the question arose how it performs as compared to the monetary metric. Indeed, if monetary measures remain the reference, then our assets index should share some common properties with them. The second chapter assesses the trends in material poverty in Ghana from the assets perspective using the Core Welfare Indicators Questionnaires Surveys (CWIQ). It then compared these trends with the monetary poverty over roughly the same period. We show that the assets index could be used and yields the same consistent results as using other welfare variable (such as income, consumption or expenditure). Therefore, using two consecutive CWIQ surveys, we find that material poverty in Ghana has decreased roughly by the same magnitude as monetary one, as found in other studies by other authors such as Coulombe and McKay (2007) using Ghanaian GLSS16 consumption data. Thus, this chapter could thus be viewed as providing the proof that the material and the monetary approaches could be equivalent. The second part of our dissertation seeks how to define and measure health and inequality in health. While the definition of health is not obvious, we propose to measure it with child mortality rates. Our main rationale in doing so is that low child mortality generates, ceteris paribus, higher life expectancy17, thus is an adequate measure of a population?s health. This may not be true in areas devastated by wars, famines, and HIV and other pandemics where child mortality could be high (in this case, the best measure should be life expectancy by age groups). Also, the reader should bear in mind that in fact, child mortality could be itself is a good indicator for measuring the (success of the) economic development level of a society as a whole (Sen, 1995), mainly because in developing countries, child mortality is highly correlated to factors linked to the level of development such as access to safe water, sanitation, vaccination coverage, access to health care, etc. - In the third chapter, we focus on measuring overall population?s health. For this, we estimate child mortality in SSA and compare it to the rest of the world. We explore the 16 Ghana Living Standard Surveys. 17 By construction, life expectancy at birth is highly correlated and sensitive to child mortality (it is based on child mortality rates for various cohorts). Lower child mortality rates lead to higher life expectancy and vice versa. determinants of child mortality using mainly a Weibull model and DHS data with socioeconomic variables18 as one of our major covariates. The use of the assets index information is to see how these quintiles behave in a multivariate regression framework of child mortality (i.e. how they affect child mortality). We find, among others, that mother?s education and access to health care and sanitation are one of the strongest predictors for child survival. Controlling for education and other factors, family?s wealth and the area of residency do not really matter for child survival in SSA, contrasting with results found elsewhere. - The fourth and last chapter answers the ultimate goal of this dissertation, that is, the scope of health inequalities in the developing world, particularly in SSA. It uses the factor analysis (FA) method of Chapter 1 to rank household according to their economic gradient status19 and then studies inequalities in various health indicators in relation with these groups. The intention is to analyze inequality rates between rich and poor for various health variables. In this chapter, we concentrate solely on inequality issues in health and health-related infrastructures and services. Mainly, we analyze inequality in access to sanitation infrastructures (water and electricity20) and various health status and access to health indicators (such as child death, child anthropometry, medically assisted delivery and vaccination coverage) using a Gini and Marginal Gini Income Elasticity approach (GIE and MGIE, henceforth) on one hand, and the Concentration Index (CI) approach on the other. Results show that, while almost all countries have made great efforts in improving coverage in, and access to, these indicators, almost all the gains have been captured by the better-offs of the society, especially in SSA. We extend the analysis to compare GIE estimates to those of CI and find consistent results yielding quite similar messages. 18 Quintiles groups derived from an assets index. 19 By grouping usually households in 5 quintiles from poorer to richer ones. 20 On the rationale of using electricity, see footnote 7 above. 3. Results and Policy Implications. As said above, the major goal in conducting this thesis research is to analyze inequality in health status, health care and health-related services using DHS data. To reach our objective, we follow two intermediate steps: - For assets poverty, results show that assets poverty and inequality are decreasing in every region of the world, including Sub-Saharan Africa. This tends to support our hypothesis that, contrary to common beliefs, Af