FILE COPY- 44- Economic Regress Concepts and Features Amar tya Sen Different criteria for measuring economic regress show varying results. Alterna- tive procedures involve parametric variations in the focal variable (such as real income, gross domestic product-GDP-or mortality rates), time stretcb (long- run decline or persistent shor-run crises such as famines), relativity (absolute decline as distinct from relative setbacks), and unit of aggregation (such as broad regions, countries, classes, genders, and income groups). For example, over the long term, absolute GDP per capita has declined in several countries, but mor- tality rates of children under age five have improved for eacb. Regress in under- five mortality can be assessed in relative terns (such as comparison with half the median improvement). There is some evidence of growing divergence in the space of these mortalities. The paper uses some of these perspectives to analyze. the nature of economic regress, to identfy some experiences in that direction, to comment on their causal antecedents, and to indicate the necessity forgoing beyond the intercountry pic- ture (despite the usefulness of that picture). The need to supplement detailed economic diagnoses (sucb as a shortage of investment) with broader investigations of the political economy of civil wars and military governments, particularly in Sub-Saharan Africa, ernerges as one of the imperatives in thisfilek It is nice to be asked to write on 'the economics of regress," but what, in fact, is regress? The language of development economics, with its incurably optimistic bias, does not make it easy to discuss the topic. Take the term "developing economy," which refers to countries that are lacking in development, whether or not they Amartya Sen is Lamont University Professor at Harvard University. He would likce w thank T. N. Srinivasan for critical comments and Stephan Klasen for research assistance and helpful suggestions. Proceedings of tue World Bank Annual Conference on Development Economics 1993 @ 1994 The lntenational Bank for Rconsrruction and Devdopment f THE WaRLD BANK 315 are a,ctually."de eloping" in any understandable sense. That usage clearly does zp?af^fdiind Le for development with its occurrence-it is like defining a { ungry person as "6eating." Regress seemns impossiblc in this framnework. But the sworld isn't quite like that. Despite unprecedented prosperity, it is replete with continued and sometimes increased destitution and expanded hun- ger, the persistent occurrence of preventable diseases and epidemics, and fre- quent outbreaks of terrible famines. These problcms are sometimes more acute than they were in the past-and that certainly is regress in some imnportant sense. To address the economics of regess is to take these problems sedeously atid to apply economic analysis to them. While economics has been camed wthk dismal sciBnce, it is arguable that the subject is not quitec dismal enouth. "Do worry, dont be happy' could be a good thime song for this session. But how should we judge economic setbacks? By what criteria is regress to be identified? Several distinct issues affect the selection of criteria, and we might as well begin by separating them out. • The focal variable. What variable should be used to evaluate progress or regress? Should we use real income or real product ksuch as GDP or gross national product-GNP--per capita), or some indicator of the quality of life, such as longevity or good health? * Time stretch. Are we to identify setbacks with a declining long-term trend, or with a sharp short-term decline? The issues raised by a slow but lasdng downturn can be quite different from those that reflect a severe temporary deterioration (as, for example, a famine). * Relativity. Is regress to be understood as a setback in absolute terms, or only as a falling behind in relation to other communities or groups? This ques- tion can be asked in the context of each varable (income, product, quality of life, and so on) and each segnent of time. • Unit selection. Are countries the right units for assessing regress? Are we mainly concerned with the average situation or with the predicament of particular groups (for example, the worst off)? A population can be parti- toned according to class, gender, age, location, occupation, relative income, and so on, and the fortunes of different groups need not move together. These divisions may cut across national boundaries; for example, the quandary of the Sahelian pastoralists is a multicountry problemn, but it involves only a relatively small minority of the population in each of the Sahelian countries (see Sen 1981; Dreze and Sen 1989). Much will undoubtedly depend on how we choose between these criteria. For example, we may select the following options: * Focal variable: large-scale premature mortality * Time stretch: quick deterioration - Relativity: absolute numbers affected * Unit: counties, or regions within them. 316 Economic Regres We would then have to pay particular attention to the extraordinary phenom- ena of famines. The economics of regress would have to take tip sudden mass starvation, illness, and decimation. This characterization will take us to- among others-Somalia and Sudan today, Ethiopia and the Sahelian countries in the recent past, Cambodia in the 1970s, Nigeria before that, and so on. Not only did those famines kill many millions; they also made the lives of a great many others unbearably miserable. This particular perspective of economic regress cannot be overlooked, if only because famines continue to occur, with distressing frequency, in the modern world. A quite different aspect of economic regress is addressed by Robert Barro and Jong-Wha Lee in their paper for this session. Although the authors go into a number of different perspectives (and provide, for example, a substantial analy- sis of the associations among income, fertility, mortality, and education), they focus on the following variables: * Focal variable: real GDP per capita * Timestretcb: long-term (mid-1960s to mid-1980s) * Relativity: comparative growth rates v Unit: aggregate country performance. This set of criteria yields quite a different identification of "losers" from what one would obtain in a mortality-centered approach. In this paper I shall try out some alternative-and often quite divergent-perspectives. Before I proceed, I would like to comment on the cogency of using a relativist perspective. If regress is to be judged in purely relative terms, it might appear that some experience of regress would always occur, since some communities would invariably achieve less than others-and indeed less than the average (unless all coincidentally have the same achievement). But -this argument is specious if it is meant to apply to all relativist standards. A suitably chosen criterion derived entirely from a relativist perspective can easily avoid this difficulty. For example, regress, in a relative sense, need not be inevitable if the chosen standard is such that a country would be diagnosed as having regressed if its absolute progress were, say, less than half the median progress level for that group of countries. It is possible that no country would fall below this line, and, depending on the distribution of achievements, there may or may not be any relative decline in terms of this yardstick. This standard will, in fact, be used later to identify regress in the reduction of morality rates. Quality of the Data The data used here come from standard international sources, including the United Nations and the World Bank, and, in one case, from research results by a team of economists, including Robert Summers and Alan Heston (1987) and, earlier, Irving Kravis. The data are of uneven quality, and the repetition of the Sen 317 same figures in scores of publication is no guarantee of reliability, since the same sources are quoted again and again. The problem with the data provided by international organizations is that they are often reconstructed rather than directly observed. This need not, in itself, make these data suspect. Indeed, in constructing the price-adjusted GDP, the numbers may well be explicitly processed with great sophistication (for example in Summers and Heston 1988, 1991). But the reconstruction is often cruder, with more guesswork than analysis, and sometimes this applies even in cases that may appear to be results of simple observation, such as age-specific mortality rates. Devised data of this type are induded in, among other sources, the World Bank's World Development Report, uNCEF's The State of the World's Children, and the uNDP's Human Development Report. There is enough arbitrariness in these figures to make the purists among us shrink at the thought of touching such data. Given the scarcity of reliable primary information, the purist position can almost lead to abstinence from empirical work altogether. It can also result in silence on some of the most pressing policy problems (even though some who are purist on empirical matters seem willing enough to rely almost entirely on pure theory to pronounce on general policy recommendations). There is a dilemma here. There is merit in being cautious, but at the same time it is not absurd to use-with adequate warning-figures that are considered by statistical organizations which specialize in gathering and reconstructing inter- national data to be the best estimates. That is the position taken in this paper, but I do want to emphasize that the status of the empirical analysis presented here is constrained by the quality of the data. Another issue involved here is that the tendency of economists to use whatever data the international organizations publish may, to some extent, contribute to reducing the institutions' incentive to improve the quality of the data. The purist may thus have a point that goes beyond the alleged worthlessness of the data. There are strategic issues both in using the data that are available at this time and in worling toward improving the general availability of data. The two tasks interconnect. The use of statistics published by international organizations (despite their possible unreliability) does, therefore, raise rather complex issues of appropriateness. The Regressors: GDP Growth and the Reduction in Mortality Table 1 gives the list of countres-sixteen in all-with negative growth of GDP per capita during 1960-85 (see Summers and Heston 1991).' They are arranged in increasing order of growth rate (negative though they all are).2 The table also pre- sents the decline of mortality rates for children under five,3 a convenient index for comparing the reduction in premature deaths (see Dr&ze and Sen 1989, ch. 10). Note that none of the countries that experienced faling real GDP per capita recorded an increase in under-five mortality rates. Indeed, something more 318 Economic Regress Table 1. Countries with Lowest Growth Rates, 1960-8S (percent) Annual growth of Annual reduction in Country GDP per capita under-fte mortality 1. Kuwait -5.82 6.32 2. Mozanmbique -1.95 1.08 3. Angola -1.75 1.41 4. Madagascar -1.66 5.15 S. Chad -1.42 1.34 6. Zambia -1.17 1.96 7. Ghana -0.91 1.51 8. Zaire -0.58 2.25 9. Central African Rep. -0.56 0.94 10. Guyana -0.44 3.26 11. Liberia -0.39 1.45 12. Mali -0.29 1.12 13. Afghanistan -0.27 0.36 14. Nicaragua -0.09 2.75 15. Somalia -O.OS 0.54 16. Benin -0.02 1.89 Sources: UICEF 1987,1993; Summers and Heson 1991. assertive can be said about the trend: no country in UNICEF'S dara set (142 countries in all-very few are left out) experienced an absolute decline in this field. In so far as we (a) take this mortality rate as the focal variable in judging regress, (b) consider long-term change over 1960-85 or, for that matter, 1960- 91 (umCEF 1993), and (c) look only at country averages, there is no evidence at all of absolute decline. The picture would be somewhat different if we chose shorter periods. During the 1980s (the latest period in the UNICEF data), under-five mortality increased in Angola, Iraq, Mozambique, and Zambia (see uNicEF 1993, table 10). Cam- bodia experienced an increase earlier (in the 1970s), and during the two decades from 1960 to 1980 under-five mortality increased 2.1 percent, on average. However, each of these counties went through a war or serious political crisis- an important issue to which I return later. .f- Table 2 presents the worst performers in under-five mortality during 1960- 85. Since none experienced an absolute increase over this long haul, I have used a relativist selection, by taking half the median value of the percentage reduction as the cutoff line. The median value of percentage decline for the countries for which such data are given by uNcIEF is about 50 percent; a cutoff line of half that value was chosen. This yielded twenty-one countries in which the percent- age of reduction in mortality was less than half the median reduction for the group. Because GDP growth data are not available for Cambodia, only twenty countries are listed. The annual percentage rates of reduction of under-five mortality are given for these countries, along with their relative regress compared with half the median Se 319 Tabic 2. Slowest Reducers of Under-Five Mortality, 1960-8S percnt) Annual rate of redetion in under-five rnortality Differencefrom Anrtualgrowdk of Country Acttal half-mnedian GDP per capita 1. Afghanistan 0.36 -0.80 -0.27 2. Somalia 0.54 -0.62 -0.05 3. Ethiopia 0.54 -0.62 0.80 4. Nigeria 0.61 -0.55 0.46 S. Rwanda 0.70 -0.46 1.37 6. South Africa 0.76 -0.40 1.61 7. Uganda 0.90 -0.26 2.57 8. Bangladesh 0.92 -0.24 0.61 9. Central African Rep. 0.94 -0.22 -0.56 10. Pakistan 0.95 -0.21 2.33 11. Sierra Lcone 0.97 -0.19 0.42 12. Swaziland 0.98 -0.18 1.97 13. Gambia, The 1.00 -0.16 0.07 14. Iran 1.01 -0.15 2.72 15. Senegal 1.03 -0.13 0.02 16. Burundi 1.04 -0.12 0.73 17. Guinea 1.05 -0.11 0.41 18. Mozanbique 1.08 -0.08 -1.9S 19. Mali 1.12 -0.04 -0.29 20. Malawi 1.13 -0.03 1.25 Souecw uNICEF 1987,1993; Summers and Heston 1991. reduction. The annual growth rates of GDP per capita are also given for purposes of comparison. Of these twenty losers in mortality reduction, only five had negative rates of growth of GDP per capita. The rankings are again quite different between GDP growcl and under-five mortality reduction among the losers in each field (see tables 1 and 2). Table 3 shows the GDP and mortality data for all the countries (a total of 118) for wHch both sets of data are available. There is something of a relation between the two orderings. Even though regressing the actual values of one on the other produces quite a weak relation (with an R-squared of 0.15), Spearman's rank correlation yields a fairly significant connection, with a coefficient of 0.53. It would be surprising had there been no relation at all, not only because there must be some causal links between success in economic growth and in mortality reduction but also because one would expect some indirect connections between them (for example, through education and urbanization). Be that as it may, the important question in the context of the present exercise is not the existence (or nonexistence) of an overall relation between the two focal variables but the connection between the -losers" that regressed in the two respective areas. Here the dissonance is apparent enough. One way of seeing 320 Economic Regress Table 3. Relationship between Growth of GDP Per Capita and Reduction in Under- Five Mortalt Selected Economies, 1960-8S Reduction in under-flve mortality Growtb of real GDP Differeneefrom per capita Anntualrate half-median Annualroie Economy Rank (prcennt) (percent) Rank (percent) Hong Kong 1 6.80 5.64 3 6.18 Barbados 2 6.68 5.52 33 2.97 Chile 3 6.57 5.41 89 0.62 Costa Rica 4 6.46 5.30 61 1.97 Kuwait 5 6.32 5.16 118 -5.82 Portuga 6 6.30 5.14 14 4.08 Spain 7 6.04 4.88 28 3.31 Japan 8 5.79 4.63 6 5.39 Singapore 9 5.55 4.39 4 5.96 China 10 5.49 4.33 17 3-97 Italy 11 5.2S 4.09 23 3.4S Madagascar 12 S.1S 3.99 115 -1.66 Luxembourg 13 5.13 3.97 50 2.39 Yugoslavia 14 5.04 3.88 19 3.69 Korea, Rep. of 1S 4.99 3.83 5 5.68 Jamaica 16 4.9S 3.79 83 1.05 Greece 17 4.95 3.79 11 4.43 Fuiland 18 4.89 3.73 27 3.37 Germany, Fed. Rep. 19 4.70 3.54 44 2.54 Austria 20 4.67 3.51 30 3.24 Canada 21 4.66 350 35 2.76 Mauritius 22 4.61 3.45 51 2.36 France 23 4.41 3.25 34 2.89 Syria 24 4.37 3.21 16 3.99 Iredand 2S 4.30 3.14 53 2.32 Malta 26 4.30 3.14 8 S.27 Switzerland 27 4.30 3.14 64 1.77 Panama 28 4.30 3.14 25 3.41 Fiji 29 4.11 2.95 68 1.59 Jordan 30 3.99 2.83 32 3.10 Malaysia 31 3.98 2.82 13 3 4.11 Iceland 32 3.97 2.81 29 3.27 Sri Lanka 33 3.98 2.82 73 1.39 Belgium 34 3.88 2.72 39 2.65 Saudi Arabia 3S 3.86 2.70 31 3.20 Thailand 36 3.83 2.67 18 3.73 Trinidad and Tobago 37 3.83 2.67 60 1.98 PapuaNcwGluinea 38 3.81 2.65 71 1.48 Myanmar 39 3.76 2.60 47 2A9 Venezuela 40 3.65 2.49 79 1.14 Denmark 41 3.60 2.44 40 2.59 Sweden 42 3.60 2.44 48 2.45 Israel 43 3.50 2.34 21 3.60 Suriname 44 3.35 2.19 36 2.74 (c.'ntidmnd) Sen . 321 Table 3 (continued) Reduction in under-fnw mortality Growth of real CDP Differencefrom per capita Annual rate half-median Annual rate Economy Rank (percent) (percent) Rank (percent) Tunisia 45 3.29 2.13 26 3.41 El Salvador 46 3.29 2.13 82 1.06 United States 47 3.29 2.13 57 2.08 Norway 48 3.28 2.12 20 3.60 Guyana 49 3.26 2.10 109 -0.44 Dominican Rep. S0 3.23 2.07 63 1.92 United Kingdom S1 3.) : 2.03 S8 2.07 Netherlands 52 3.10 1.94 45 2.51 Australia 53 3.07 1.91 S6 2.22 Cyprus S4 2.96 1.80 9 4.69 Algeria SS 2.88 1.72 49 2.45 Turkey 56 2.88 1.72 41 2.58 Morocco 57 2.81 1.65 24 3.44 Guatemala 58 2.77 1.61 81 1.07 Nicaragua 59 2.7S 1.59 105 -0.09 Ecuador 60 2.73 1.57 38 2.72 Honduras 61 L70 1.54 80 1.10 Brazil 62 2.67 1.51 12 4.23 Egypt 63 2.56 1.40 7 5.29 C6ve d'Ivoire 64 2.56 1.40 70 1.48 Togo 65 2.55 1.39 65 1.75 Mexico 66 2.S2 1.36 46 2.50 New Zealand 67 2.45 1.29 77 1.31 Colombia 68 2.34 1.18 55 2.30 Peru 69 2.33 1.17 85 1.00 Congo 70 2.33 1.17 10 4.43 Uruguay 71 2.28 1.12 100 0.04 Zaire 72 2.25 1.09 111 -0.58 Argentina 73 2.21 1.05 91 0.60 Cape Verde 74 2.16 1.00 59 2.04 Botswana 75 2.12 0 0.96 1 6.66 Indoncsia . 76 2.11 0.95 15 4.04a Iraq 77 2.08 0.92 75 1.35 Kenya 78 2.03 0.87 72 1.47 Cameroon 79 2.02 0.86 22 3.55 Philippines 80 1.96 0.80 67 1.59 Zambia 81 1.96 0.80 113 -1.17 Paraguay 82 1.89 0.73 43 2.56 Benin 83 1.89 0.73 .103 -0.02 Sudan 84 1.77 0.61 92 0.59 Bolivia 85 1.61 0.45 76 1.34 Haiti -86 1.61 0.45 102 0.01 Zimbabwe 87 1.60 0.44 69 1.55 India 88 1.59 0.43 94 0.46 Ghana 89 1.51 0.3S 112 -0.91 Lesotho 90 1.50 0.34 2 6.24 322 Economic Regress Table 3 (continued) Reduction in under.fJwe mortality Growth of real CDP Differencefrom per capita Annual rat haolf-nedian Annual rate Economy Rank (percent) (percent) Rank (percent) Guinea-Bissau 91 1.47 0.31 84 1.01 Nepal 92 1.47 0.31 86 0.81 Liberia 93 1.4S 0.29 108 -0.39 Mauritania 94 1.45 0.29 97 0.38 Angola 95 1.41 0.2S 116 -1.75 Chad 96 1.34 0.18 114 -1.42 Tanzania 97 1.22 0.06 54 2.31 Niger 98 1.21 0.0S 98 0.22 Malawi 99 1.13 -0.03 78 1.25 Mali 100 1.12 -0.04 107 -0.29 Mozambique 101 1.08 -0.08 117 -1.95 Guinea 102 1.05 -0.11 96 0.41 Burundi 103 1.04 -0.12 88 0.73 Senegal 104 1.03 -0.13 101 0.02 Iran 105 1.01 -0.15 37 2.72 Gambia, The 106 1.00 -0.16 99 0.07 Swaziland 107 0.98 -0.18 62 1.97 Sierra Leone 108 0.97 -0.19 95 0.42 Pakistan 109 0.95 -0.21 52 2.33 Central African Rep. 110 034 -0.22 110 -0.56 Bangladesh 111 0.92 -0.24 90 0.61 Uanda - 112 0.90 -0.26 42 2.57 South Africa 113 0.76 -0.40 66 1.61 Rwanda 114 0.70 -0.46 74 1.37 Nigeria 115 0.61 -0.55 93 0.46 Ethiopia 116 0.54 -0.62 87 0.80 Somalia 117 0.54 -0.62 104 -0.05 Afghanistan 118 0.36 -0.80 106 -0.27 a. Rcfers to 1962-8s. Sources: UNICEF 1987,1993; Sunmers and HMesn 1991. that divergence is to note that if we list growth losers and mortality losers (talcing countries for which both sets of data are available), there are thirty-one countries in all, and only five lose in both areas. Indeed, some of the losers in one field record quite a distinguished performance in the other. Some of the relative losers in mortality reduction (Iran and Pakistn, for example) are among the better performers in GDP growth, and two of the four biggest losers in growth (Kuwait and Madagascar) are among the top performers in- the reduction of under-five mortality. Regressive Connections On the basis of the preceding discussion, it appears that it does make a differ- ence what focal variable we use to characterize regress. In particular, it matters Sen 323 how we respond to the question: regress of what? There are good reasons to be concerned about both real income and mortality rates, since both are important to us. But they are not important for the same reasons, and they point the finger at rather different countries. This analysis can be seen in some ways as complementary to-but symmetri- cal with-an earlier work done jointly with Jean Dr&ze (Dreze and Sen 1989, part iii) that focused on winners rather than on losers. We argued that high growth of GDP (or of GNP) is helpful in reducing mortality, not only because people are better able to afford good nutrition and health care but also because the resources generated can be used, if so chosen, to expand social support for education and public health, especially for the chronically deprived. In our list of ten winners in reducing under-five mortality, it turned out that half of them were indeed countries with high growth of real per capita income. We contrasted that "'growth-mediated" route to social progress with the expe- rience of the remaining five winners, which had seen little growth in the period in question but had nevertheless managed to reduce mortality quite dramati- cally. This progress was made possible largely through public action, including the expansion of epidemiological programs, general health care, basic educa- tion, and, in some cases, nutritional support. Much turned on the nature of iternal politics and the role of the public, induding its cooperative as well as productively adversarial functions (in demanding action, censuring failure, and, in general, making governments responsible and sensitive).4 We argued further that one economic reason why even very poor countries that are politically inclined to reduce mortality and expand basic education can afford to do so is that the services involved are labor-intensive and labor is relatively cheap in poor countries. For example, China, Sri Lanka, and the state of Kerala in India have been able to provide extensive health care and educa- tional services despite modest public budgets. By contrast, many countries have managed to raise GNP or GDP quite rapidly but have not used the resources generated to advance health and education, and as a result the impact on mortal- ity has been rather minimal. On balance, economic growth can certainly help reduce mortality and enhance eucation, but that help is not invariably utilized, and it is not the only possiole route. Mortality rates may remain high despite economic growth and may decline rapidly even if growth falters. And just as there are countries that have experienced "growth-mediated" social progress, with both economic growth and reductions in mortality, such as the Republic of Korea, Hong Kong, and Singapore, so too there are countries with failures in both fields, such as Somalia, Mozambique, and Mali, and there is some evidence that each failure has fed the other (see Barro and Lee, in this volume; Dreze and Sen 1989, part n; and the literature cited). Anand and Ravallion (1993) argue in a recent paper that even though there is an observed-and much discussed-association between life expectancy and per capita real income, the connection "vanishes once one controls for incidence of 324 Economic Regress poverty and public spending on health." This result can be interpreted not as a demonstration that growth of real income does not matter in reducing mortality but as a pointer to the main ways in which increases in real income favorably influence the expansion of longevity (in so far as they do influence it). The argument is that the enhancement of average national income tends to work through (a) increases in the incomes specifically of people below the poverty line and (b) increases in the funds available for expanding social services (including health care and education). To the extent that increased income would have these effects, a reduction in mortality could certainly be expected. But the increased income could also be spent in other ways, without much impact on mortality. This conditionality may explain why the winners in economic growth and mortality reduction overlap somewhat but not entirely. A similar reasoning would help us understand why there might be some overlap of losers in these two fields but would also make it quite plausible that each field might have its own home-grown losers, without their being losers all around ("champion losers," as it were). Other relations have also emerged on the basis of recent research (see Barro and Lee, in this volume).5 There are a lot of explanatory insights in these works-perhaps even an oversupply of explanation (a little like the plethora of equilibria in the theory of finitely repeated games, making it hard to choose between them). There is certainly scope for furither scrutiny. I shall come back again to this general question, but before that I would like to comment on a particular aspect of regress in mortality compared with regress in growth of income. Convergence? There is an extensive literature on whether the differences among countries in real income and economic growth are narrowing. I shall not investigate this issue (see, however, Barro and Lee in this volume). I shall, however, comment on a similar issue in mortality reduction. Table 4 shows the decline in under-five mortality rates during 1960-91 for all the countries included in UNICEF'S tables. The countries with higher initial under-five mortality have tended to record a larger absolute reduction.6 This is not remarkable, since a country with an already low value can scarcely have as large an absolute reduction. What is more interesting is the relation between levels of mortality and the annual percentage reduction in those figures. Here the relationship is quite the reverse: countries with high initial levels of mortality were relatively less success- ful in proportionate terms in reducing these deaths.7 There is a relative regress in this particular sphere of performance. Even though aggregation does not, in general, add to articulation, it is per- haps still useful to see how the relative picture stands if the countries are classi- sen 325 Table 4. Reduction in Under-Five Mortality, 1960-91 Deaths ,pr Arn;ual rate 199J 1.000 Ji-v bir/bs nf rrdrtr/ior Econaomy rnk 1960 199 (porconrt) Angola I 34S 292 0.66 Mozambique 2 331 292 0.50 Afghanistan 3 360 257 1.34 Sierra Leone 4 385 253 1.67 Guinea-Bissau S 336 242 1.30 Guinea 6 337 234 1.43 Gambia, The 7 375 234 1.87 Malawi 8 365 228 1.86 Mali 9 400 22S 2.28 Niger 10 321 218 1.54 Chad 11 325 213 1.68 Ethiopia 12 294 212 1.30 Somalia 13 294 211 1.32 Mauritania 14 321 209 1.70 Burkina Faso 15 363 206 2.24 Bhutan 16 324 205 1.81 Zambia 17 220 200 0.38 Liberia 18 310 200 1.74 Uganda 19 223 190 0.64 Rwanda 20 255 189 1.19 Nigeria 21 212 188 0.48 Cambodia 22 217 188 0.57 Senegal 23 299 182 1.97 Yemen 24 378 182 2.88 Burundi 2S 260 181 1.44 Zaire 26 300 180 2.02 Central African Rep. 27 294 180 1.94 Tanzania 28 249 178 1.33 Madagascar 29 364 173 2.93 Sudan 30 292 169 2.16 Cabon 31 287 161 2.29 Benin 32 310 149 2.89 Laos 33 233 148 1.80 Nepal 34 298 147 2.79 Togo 35 305 144 2.96 Iraq 36 171 143 0.71 Lesotho 37 210 137 1.69 Haiti 38 270 137 2.68 Ghana 39 224 137 1.95 Pakistan 40 221 134 1.98 Bangladesh 41 247 133 2.45 Comoros 42 233 133 2.22 Cbte d'lvoire 43 300 127 3.38 India 44 236 126 2.48 Cameroon 45 270 126 3.00 Bolivia 46 282 126 3.17 326 &onomic Regress Table 4 (candnued) Deaths per Annual rate 1991 1,000 live births of reduction Economy rank 1960 1991 (percent) Namibia 47 248 120 2.36 Myanmar 48 237 117 2.78 Swaziland 49 233 113 2.85 Congo S0 220 110 2.73 Libya 51 269 108 3.S8 Peru 52 240 97 3.56 Guatemala 53 220 92 3.43 Morocco 54 26S 91 4.19 Turkey 55 216 89 3.48 Zimbabwe 56 181 88 2.84 Indonesia S7 215 86 3.60 Botswana 58 169 85 2.71 Egypt 59 260 85 4.37 Ecuador 60 184 82 3.18 Mongolia 61 185 82 3.20 Nicaragua 62 209 81 3.72 Papua New Guinea 63 248 79 4.47 Dominican Rep. 64 200 76 3.80 Kenya 65 202 75 3.89 Honduras 66 230 73 4.49 Sout Africa 67 126 72 2.21 Guyana 68 126 69 2.38 Brazil 69 179 67 3.85 El Salvador 70 210 67 4.47 Ian 71 233 62 5.16 Algeria 72 243 61 5.38 Cape Verde 73 164 61 3.88 Paraguay 74 103 59 2.20 Tunisia 75 254 S8 5.74 VietNam 76 219 52 5.S9 Syria 77 217 47 5.94 Jordan 78 180 46 5.31 Lebanon 79 91 46 2.69 Philippines 80 128 46 4.01 Venezuela 81 114 43 3.82 Saudi Arabia 82 292 43 7.38 Oman 83 378 42 8.41 Suriname 84 96 37 3.74 Mexico ..5 138 37 5.13 Romania b6 82 34 3.46 Korea, Dem. Rcp. 87 120 34 4.92 Thailand 88 146 31 5.77 U.S.S.R. 89 53 31 2.12 Albania 90 151 31 6.14 Panama 91 1OS 30 4.89 (contind) Sen 327 Table 4 (continued) Deaths per Annual rate 1991 l,O00hlie births of reduction Economy rank 1960 1991 (percent) Fiji 92 97 30 4.59 United Arab Emirates 93 239 29 8.09 Mauritius 94 104 28 S.11 China 95 205 27 7.79 Uruguay 96 S7 24 3.40 Argentina 97 70 24 4.19 Trinidad and Tobago 98 69 23 4.30 Yugoslavia 99 113 22 6.34 Bulgaria 100 70 21 4.70 Chile 101 142 21 7.36 Sri Lanka 102 130 21 7.03 Colombia 103 130 21 7.03 Malaysia 104 105 20 6.42 Jamaica 1OS 89 19 S.99 Costa Rica 106 122 18 7.37 Bahrain 107 208 18 9.32 Malta 108 42 17 3.55 Hungary 109 57 17 4.72 Poland 110 70 17 5.50 Kuwait 111 128 17 7.76 Cuba 112 91 14 7.21 Czechoslovakia 113 33 13 3.66 Israel 114 39 12 4.61 Barbados 115 90 12 7.74 Portugal 116 112 12 8.55 Cyprus 117 36 11 4.63 United States 118 30 11 3.93 Greece 119 64 11 6.80 Irland 120 36 10 4.99 Singapore 121 S0 10 6.23 Belgium 122 35 10 4.89 Australia 123 24 10 3.44 Italy 124 50 10 6.23 NewZealand 125 26 10 3.75 Kora, Rcp. of 126 126 10 9.64 Canada 127 33 9 5.06 Luxembourg 128 41 9 5.89 Germany, Fed. Rep. 129 40 9 5.79 Denmark 130 25 9 4.00 Switzerland 131 27 9 4.30 France 132 34 9 5.18 United Kingdom 133 27 9 4.30 Austria 134 43 9 6.06 Spain 135 57 9 7.12 Hong Kong 136 64 8 7.98 Netherlands 137 22 8 3.97 328 -conomic Regress Table 4 (continued) Deaths per Annualrate 19n1 1. 000 liue births ofraductin Economy rank 1960 1991 (pecent) Norway 138 23 8 4.14 Iceland 139 22 7 4.48 Finland 140 28 7 S.39 Japan 141 40 6 7.31 Sweden 142 20 5 S.39 Sources: UNICEF 1987,1993. fled according to absolute levels of under-five mortality in 1991 (see UNICEF 1993). During 1960-80 the "very high mortality" countries achieved an average reduction rate of 1.1 percent a year, whereas those in the next, "high," category recorded 1.6 percent, the "middle" countries 4.4 percent, and the already privi- leged "low" countries a full S percent reduction a year. (Very high mortality is defined as 140 per thousand, high mortality as 71-140 per thousand, middle mortality as 21-70 per thousand, and low mortality as 20 or less per thousand.) This monotonic relative regressivity is almost duplicated in 1980-91, with the "'very high" group recording an annual reduction of 1.0 percent, the "high" group 3.1 percent, and the "middle" and "Iow" categories 5.8 percent and 4.1 percent, respectively. The worsening relative predicament of those who are already worst off is clear enough. Civil War and Militarization in Sub-Saharan Africa It may be useful to look at under-five mortality rates by region to draw attention to the one region in which regress is more of a reality today than anywhere else. The reduction of under-five mortality in percentage terms in Sub-Saharan Africa was less than half that achieved by any other region in the later period 1980-91 and was consistently less throughout the entire span.8 Any serious study of the economics of regress must, to a great extent, focus on this region. In fact, one of the common features of both lists of losers (tables 1 and 2)-in tenns of GDP growth and mortality reduction-is the preponderance of countries in Sub-Saharan Africa. Twelve of the sixteen growth regressors and sixteen of the twenty mortality regressors are in Sub-Saharan Africa (four counties in this region 'lose" . -.oth fields, of a total of five in the world). There is a danger of losing sight of the forest in our concentration on the trees. Although examination of regress on the basis of individual countries is impor- tant, it is also necessary to keep the bigger regional picture in mind. In that picture Sub-Saharan Africa stands out sharply enough. In terms of the standard growth variables identified in the literature, the causal antecedents of this low performance are easy to see. Consider, for exam- sen 329 pie, domestic investment. Gross domestic investment increased by 4.6 percent a year in South Asia and by 10 percent a year in East Asia and the Pacific during 1980-90, but it fell 4.3 percent annually in Sub-Saharan Africa (World Bank 1992, table 8 in the World Development Indicators). There are other connections-some more obvious than others-involving indebtedness, drought, high fertility, and so on, which have all been covered in the literature in this field. An aficionado of numerical gloom can sift through a lot of telltale information on the decline of Sub-Saharan Africa in the World Bank's World Development Reports, the UNDP's Human Development Reports, umla's The State of the World's Children, and other annals of international official history. But these detailed economic data need to be supplemented by broader investiga- tions of the general political economy. No major region has been as affected by civil wars, military governments, and other destructive political forces. In the list of worst performers identified earlier in this essay, the association is hard to miss. If, for example, we pick the five worst performers in economic growth and the five worst in mortality reduction out of this group of ten (see table 5), we get eight that are Sub-Saharan coun- tries, and six of these have had major civil wars-Somalia, Ethiopia, Nigeria, Mozambique, Angola, and Chad. Of the two countries outside Sub-Saharan Africa, Kuwait is obviously a peculiar case; its low GDP growth is offset by one of the finest records in reducing infant mortality. The remaining country is Afghanistan, whose unenviable position as the worst performer in mortality reduction is not unrelated to civil war.9 Table S. Growth and Mortality Reductionfor Top Ten Regressors, 1960-85 Unde-fie mortality Real GDPper capita 198S Annual Annual value differncne growth (per 1,000 from hal- Presence Of Dolaars, raIe live median Pcc of counr 1985 (Percent) births) (percent) Civil tar Famine Slowest mortaity reducers Afghanistan 714 -0.27 329 -0.80 Yes No Somalia 843 -0.05 257 -0.62 Yes Yes Ethiopia 325 0.80 257 -0.62 Yes Yes Nigeria 1,066 0.61 182 -0.55 Yes Yes Rwanda 731 0.70 214 -0.46 No No Slowestgrowers Kuwait 12,684 -5.82 25 5.16 No - No Mozambique 831 -1.95 252 -0.08 Yes Yes Angola 840 -1.75 242 0.25 Yes No Madagascar 677 -1S5 97 3.99 No No Chad 519 -1.42 232 0.18 Yes Yes Sources: UMC 1987,1993; World Bank 1987; Summes and Hesron 1991. 330 Economic Regress It is not my contention that we can dose our eyes to complex economic features in understanding economic regress in Sub-Saharan Africa. It is, rater, that there is some need to associate the region's predicament with its overall experience of wars and military governments. In fact, the detailed intercountry picture (loud and clear as it is) does not do full justice to the consequences of war and militarism, since the impact often extends beyond the countries that are direcdy involved, reducing regional trade and, in neighboring countries, causing political instability, weakening the climate for economic investment, and induc- ing diversions of economic resources to military purposes. The association of wars with famines has been discussed elsewhere (see Drbze and Sen 1989; Sen 1991), and that causal connection also has implications for long-run regress. Civil wars not only affect capital and output; they also disrupt trade and commerce, discourage investments, and, last but not least, tend to increase the power of the military and to distance the political leadership from the economic hardships of the population. The encouragement of militarization and dictatorships in that context has a particular role in reducing the political incentive to undertake economic and social development. That general connec- tion must, I believe, be kept in view in assessing the overwhelming presence of economic regress in Sub-Saharan Africa. Condusions There are several points that emerge from this paper. First, economic regress can be seen in terms of various criteria that provide different views of decline. Of partiailar relevance are the parametric variations in the focal variable, time stretch, relativity, and unit of aggregation. The literature in the field can be discriminatingly classified in terms of these variations. The issue of appropriate focus is a substantial one, as is the question of the reliability of the data pre- sented by leading international organizations. Second, although many countres retrogressed in absolute terms in real GDP per capita during 1960-85, none seem to have regressed-in absolute terms-in under-five mortality. This is true also for the rather longer period from 1960 to 1991. But the relative performances are extremely variable, and given the gen- eral improvement in medical opportunities across the world, the relative failure of some countries can be seen as quite serious. By using half the median value of percentage mortality reduction as the point of comparison, the countries that regressed (in this relative perspective) can be identified. Third, there is overlap as well as divergence in the respective patterns of regress in real income and mortality. The various causal interconnections explored in the literature indicate some fruitful lines of investigation of this partal concordance, although the connections need further assessment. Fourth, there is much evidence against sconvergencen in the reduction of under-five mortlity; rates decined in the worst-performing countries by smaller percentages. There is, in this sense, a regress of some importance, which sen 331 demands a counteracting response. Perhaps because of the general success across the world in reducing absolute mortality rates, this extensive regress has failed to receive the professional attention it would seem to deserve. Fifth, Sub-Saharan Africa stands out as the most "regressive" region in terms of both real income and mortality rates. While explanations of this predicament are plentiful and involve many economic variables, it is important not to lose sight of the general impact of the civil wars and military govemments that plague the region. The role of these influences on economic regress can be identified in part in terms of intercountry comparisons, but often the impact extends beyond the countries directly involved, adversely affecting trade, invest- ment, and the economies of neighboring countries. Finally, the problems of famine and hunger, which have been discussed else- where (including in Drtze and Sen 1989) relate to long-mn regress. The influ- ence of wars and military governments that dominates Sub-Saharan Africa has much to do with both types of failure. An economic analysis of regress cannot afford to overlook that political connection. Notes 1. The calculations here go beyond Summers and Heston (1991), and I am grateful to the authors for access to their computerized information. The series taken is indexed at 1960 values and uses the Laspeyres weights. 2. In their paper Barro and Lee have dealt with 19658S data from the same source (smong others) ad have identified twenty countries with negative growth rates. Of die sixteen countries in the negadve growth categoqry in my list for 1960-85, fourteen are also in the Barro and Lee list for 196S85. 3. The 1960 figures come from UNICEF (1993). To make comparison with the growth of GDP over the same period (1960-85) for both, the other end is taken to be 1985, for which data are obtained from uwcEF (1987). 4. See Dreze and Sen (1989), chs. 10-13, and the extensive literature dted there, particularly Streeten and others (1981); Halstead, Walsh, and Warren (1985); Stewart (1985); Caldwevl (1986); Behrman and Deolalikar(1988); and Griffin and Knight (1990). S. See also Jolly and Comni (1984); Halstead, Walsh, and Warren (1985); Caldwel (1986); Cornia, Jolly and Stewart (1987); Schultz (1987); Behrman and Deolalikar (1988); Bel and Reich (1988); Dr&ze and Sen (1989, 1990); Grffin and Knight (1990); Chen, Kleinman, and Ware (1992); Coniia, van der Hoeven, and Mlanawire (1992); Stewart, Lall, and Wangwe (1992); and Osmani (1992). 6. Relating absolute reduction to the absolute level of under-five mortlity yields a positive coefficient of 0.3813, with a standard error of 0.0308, and a correlation coefficieat of 0.72 for these 142 observations. 7. The coefficient in this relation is negative, in particular -0.000094, which is not as slight as it might first appear, since percentage figures are small (the median value of pentage reduction was 3.8 percent), whereas the absolute numbers of deaths per thousand are large (the median value is aroud 233 in 1960). The standard error is 0.000014, and the correlation coicient is 0.49. 8. For both 1960-80 and 1980-91, Sub-Saharan Africa's annual reduction rates werc 1.2 percent, whereas the figures for South Asia were 1.4 for 1960480 and 2.8 percent for 1980-91, for Latin-America and the Caribbean 3.0 and 4.0 percent, for the Middle East and North Airica 2.6 and 3.7 percent, and for East Asia and the Pacific 4.2 and 5.5 percent. 9. There is, in fact, one country for which mortlity-but not growth-figures are reported and which thus is not included in the table. That country is Cambodia, which allegedly did even worse than Afghanistm in the field of mortality reduction. The association with civil wars and military governments is not unobvious in this case either. 332 Economic Regress References Anand, Sudhir, and Martin Ravallion. 1993. "Human Devclopmcnt in Poor Countries: On the Role of Private Incomes and Public Services," Journal of Economic Perspectives 7 (Winter): 133-50. Bchrman, J. R., and Anil B. Deolalikar. 1983. "Health and Nutrition." In Hollis Chenery and T. N. Srinivasan, eds., Handbook of Development Economics. Amsterdam: North-Holland. Bell, 0. E., and M. R. Reich, eds. 1988. Health, Nutrition and Economic Crises. Dover, Mass.: Auburn House. Caldwell, J. C. 1986. 'Routes to Low Mortality in Poor Countries." Population and Development Review 12 June): 171-220. Chen, L. C., A. Kleinman, and N. C. Ware. 1992. Advancing Healtb in Developing Counties. New York: Auburn House. Chenery, Hollis, and T. N. Srinivasan, eds. 1988. Handbook of Development Economics. Amsterdam: North-Holland. f CoMia, C. A., Rolph van der Hoeven, and Thondika Mkandawire, eds. 1992. Africa's Recowery in the 1 990s: From Stagnation and Adjustment to Hwnan Development. New York: St. Martin's Press. Cornia, G. A., Richard jolly, and Frances Stewart. 1987. Adjustment wtth a Human Face. Oxford, U.K.: Carendon Press. Drfe,Jean, and Amartya K. Sen. 1989. Hunger and Public Action. Oxford, U.K.: Carendon Press. -. eds. 1990. The Political Economy of Hunger. Oxford, U.K.: Carendon Press. Griffin, Keith, and John Knight, eds. 1990. Human Development and the Intenational Devlopment Strategyfor the 1990s. London: Macmillan. Haistrad, S. B.,J. A.Walsh, and K. S. Warren. 1985. Good Healh at Low Cst. New YorLh Rockefdler Foundation. Jolly, Richard, and G. A. Comia, eds. 1984. The Impaa of World Recession on Cbildren. Oxford, U.K.: Pergamon. Osmani, S. R. 1992. Nutrition and Poverty. Oxford, U.K.: Clarendon Press. Schultz, T. Paul. 1987. Education, westment, and Returns in Economic Development New Haven, Conn.: Yale University Press. Sen, Amarrya K. 1981. Poverty and Famines. Oxford, U.K.: Clarendon Press. . 1992. "Wars and Famines." In Walter Isard and Charles H. Anderton, eds., Economics of Arms Reduction and the Peace Process. Amsterdam: North-Holland. Stewart, Frances. 1985. Planning to Meet Basic Needs. London: Macmillan. Stewart, Frances, S. Lall, and S. Wangwe, eds. 1992. Alternative DevdopmentStrategies in Sub-Snharan Africa. London: Macmillan. Streeten, Paul, Shahid javed Burki, Mahbub ul Haq, Norman Hicks, and Frances Stewart, eds. 1981. First Things First Meeting Basic Needs in Developing Countries. New York: Oxford University Press. Summers, Robert, and Alan Heston. 1988. "A New Set of International Comparisons of Real Produas and Prices: Estimates for 130 Countries, 1950-85." Review of Income and Wealth Series 34 (1) (March): 1-25. - . 1991. 4The Penn World Table (MarkS): An Expanded Setof International Comparisons, 1955- 1983." QuarterlyJournal ofEconomics 106 (May): 32768. uNDP (United Nations Development Progranme). Various issues. Human Development Report. New York. UNICEF (United Nations Children's Fund). Various issues. The State of the Worli's Childrn. Oxford, U.K.: Oxford University Press. World Bank. Various issues. World Development Report. New York. Oxford Universiy Press. Sen 333 COMMENT ON "ECONOMIC REGRESS: CONCEPrS AND FEATURES," BY SEN Nora Lustig professor Sen makes the point that countries which experienced economic regress according to one variable-growth of gross domestic product (GDP)-do not necesarily belong in the losers column when another vari- able, under-five infant mortality, is used instead. That is, the countries with poor GDP performance are not ncessarily the same as those with high under-five infant mortality. According to Sen's findings, of the thirty-two countries identi- fied as losers under one of the two criteria, only five lose in-both areas. Thus, Sen concludes, the choice of the focal variable used to characterize regess does make a difference. Although this is certainly possible in general, in the particular case of GDP performance and under-five infant mortality the degree of divergence between the lists of growth losers and mortality may be sensitive to a number of assumptions that would be worth exploring. First, how robust are these results to changes in time span? For example, if we compare tables 4 and 5 in the Sen paper using the 1960-91 figures for under-five infant mortality, Mozambique and Angola fall into the "slowest mortality reducerse category and Afghanistan, Somalia, and Rwanda stand to move out of that category because their rate of improvement was substantially higher in 1985-91 than in 1960-85. The question is, would the small overlap between bad performers under both criteria change if different time spans were used? Second, how robust are the results when the definiton of "relative perfor- mance' is changed? Sen uses half the median value of the percentage reduction of under-five mortality during 1960-85 as the cutoff line. Alternatively, we could classify countries in groups according to income per capita and regress infant mortality against that measure; those countries whose actual performance is worse than the predicted performance could be classified as doing relatively poorly in terms of infant mortality. Then one could proceed to check whether the overlap between bad performers in terms of GDP and under-five infant mor- taliy is stronger, weaker, or the same as with the criteron used by Sen. Nora Lustig is senior felow in foreign policy studies at the Broolings Institution. Proceedings of the World Bank Annul Conference on Development Economics 1993 © 1994 The International Bank for Reconstruction and Development THE WORLD M 335 Third, how do the results change when countries are classified according to initial income? It seems reasonable to expect that the starting point should make a difference. For example, Kuwait's bad performance in terms of GDP but not in terms of under-five infant mortality should not be surprising. Kuwait's per capita income is ten (or more) times higher than that of the Sub-Saharan African countries. A decline in per capita income in a rich country should in principle have less effect on the variables that affect under-five jLfant mortality than it would in a poor country, or perhaps no effect at all. The overlap between bad and good performers may be higher if one controls for different initial conditions. Let us assume that the small overlav between losers on the growth list and the mortality list is robust. As Sen and others have noted, under-five infant mortal- ity may not be very sensitive to growth of GDP because of the vast -technological changes in the prevention of avoidable deaths. For example, mass inoculation and oral rehydration have reduced infant and child mortality at a relatively low cost, and these practices can be continued even if GDP falls and governments cut public spending, as long as the initial dissemination of know-how has occurred. -I Some of the most important measures for preventing infant deaths have a large component of inerdia and are resilient to dedines in GDP. One crucial policy implication is the overriding importance of setting preventive mechanisms in place and expanding them to currdntly deprived regions. This should come as no news, but it does not hurt to stress it. T-he findings in Sen's paper bring out an important point: those countries with the worst initial levels of under-five infant mortality are the least effective in reducing it. This result is troubling not only for humanitarian reasons but because it is the converse of what one would expect: that very high mortality rates should be easier to reduce. It is important to establish whether the problem reflects lack of access to inexpensive preventive technologies or exogenous fac- tors, such as war and famine, that affect certain regions or countries. If the problem is lack of access to technology, multilateral development institutions can facilitate access to this technology. In the case of famine, they can develop schemes that increase food security as a preventive strtegy and can provide relief to famine-stricken areas. Clearly, the most difficult problem to tackle is war and overall political and military unrest-unfortunately, under these cir- cumstances there is much less that the outside world can do. Finally, trning to the possible link between economic regress and the role of war and other exogenous factors, Sen finds that economic regress is more fre- quent in Sub-Saharan African countries that are characteristically ridden by wars and political unrest-a finding that is supported in a recent World Bank paper on the role of policy in explaining growth performance (Easterly and others 1993).1 According to that paper, change in growth perfonrance is not explained primarily by policies or rates of factor accumulation. Rather, much of the variance in growth rates can be explained by such shocks as wars, debt crises, external transfers, and changes in tenns of trade. Of the shock variables, 336 Comment changes in terms of trade were the most important. War casualties were signifi- cant in the 1980s in the African context, and although the authors do not report results for individual countries, they mention that the civil war in Uganda inay have reduced growth in the 1970s by 3 percentage points a year. The fact that random shocks play such a fundamental role in explaining performance may be disappointing, particularly for an institution such as the World Bank, where policy advice is the bread and butter of everyday activity. If economic progress is so susceptible to random factors, a substantial part of the policy recommendations should come in the form of some sort of mechanism that permits countries to hedge against bad times. It is conceivable that the World Bank could help devise insurance schemes that would operate to smooth the effects of fluctuating income in individual countries. Such schemes could be vastly important in reducing the probability of economic regress, independently of the variable used to measure it. Reference Easterly, William, Michad Kremer, Lance Pritchett, and Lawrence Summers. Forthcoming. "Good Policy or Good Luck? Country Growth Performance and Temporary Shocks." Jomrnal of Monetary Economics. Lustig 337 COMMENT ON "ECONOMIC REGRESS: CONCEPTS AND FEATURES," BY SEN Nurul Islam J agree with Professor Sen's contention that economic regress can be analyzed in terms of different criteria that will, in turn, provide different views of economic setbacks. My comments are primarily supplementary and will also raise a few unsettled questions and issues that require furither analysis. Sen finds that the worst performers in the growth of gross domestic product (Gpp) per capita were different from the worst performers in the rate of reduction in mortality of children under five years old. Further, among the countries that experienced negative growth rates during 1960-85, there was no correlation between the rate of decline in GDP per capita and the rate of reduction in mortality; similarly, among the countries with the worst records of infant mortality, there was no correlation between the two criteria. Among the slowest-growing countries that at the same time did reduce mortality, a few achieved a rate of reduction that was higher than the median for all 118 countries. What accounts for this improvement in the face of actual declines in per capita income? This question is of particular relevance in the context of ongoing eco- nomic policy reforms or structura adjustment programs that often depress income, at least in the short or medium term. Several tentative answers come to mind. One is that some countries, despite a decline in per capita income, had high absolute levels of per capita income and a low level of poverty. They probably could afford-and in fact might have had-a high level of public expenditure on health or a high percentage of population with access to health services. In fact, it appears tat this was the case in such counties as Ghana, Kuwait, Nicaragua, and Zambia. There were also a few countries with low per capita income or below-average access to health services and below-average expenditures on health (such as Afghanistan, Central African Republic, Mozam- bique, and Somalia) that also managed to reduce mortality rates, although more Nurul lslam is senior policy adviser, International Food Policy Research Institute, Washington, D.C. Proceedings of the World Bank Anal Conference on Development Eonomics 1993 @ 1994 The Intnational Bank for Reconstruction and Development t THE WORLD BANK 339 Table 1. Countries with Lowest Growth Rates, 1960-SS Annual Percentage of Publie rate of popuknion expenditure red"ction with access on cinalcm as of utnder-fie to beatth Per capita income Percentage of mnortality seruices, (dollars) GNP Country (percent) 19854I7 1976 1985 1960 1985 1. Kuwait 6.32 100 15,480 14,480 - 2.9 2. Mozambique 1.08 39 170 160 - 1.8 3. Angola 1.41 30 330 - - 1.0 4. Madagascar 5.15 56 200 240 1.4 1.8 S. Chad 1.34 30 120 - 0.5 0.6 6. Zambia 1.96 75 440 390 1.0 2.1 7. Ghana 1.51 60 580 380 1.1 0.3 8. Zaire 2.25 26 140 170 - 0.8 9. Central African Rep. 0.94 45 230 260 1.3 1.2 10. Guyana 3.26 - - - - - 11. Liberia 1.45 39 450 470 0.8 1.5 12. Mali 1.12 15 100 150 1.0 0.7 13. Afghanistan 0.36 29 160 - - - 14. Nicaragua 2.75 83 750 770 0.4 6.6 15. Somalia 0.54 27 110 28 0.6 0.2 16. Benin 1.89 18 130 260 1.5 0.8 - Not avaitable. Note: The average pubic expenditure on health as a share of GNr for all developing countries in 1986 is 1.4 percet. Sources: World Bank 1977,1987; UNDP 1990. slowly. Of course, there are several countries for which this pattern of explana- tion does not hold (see table 1). If we look at the countries with the slowest reductions in mortality (see table 2), we find that as late as 1985 the majority had per capita income of less than $300; in only two countries-Iran and South Africa-was per capita income higher than $1,000. Two more, Pakistan and Sierra Leone, recorded per capita incomes of $380 and $350, respectively, while one country, Nigeria, had a per capita income of $800. Furthennore, in only three countries was the percentage of population with access to health services higher than the average for all developing countries (that is, 61 percent during 1985-87). Some countries were significantly below the average; in 1985 seventeen of the twenty countries spent less on health, as a percentage of gross national product (GNP), than the average for developing countries. If the worst-performing countries,.as ranked by GDP growth, are so different from the worst-performing countries in mortality reduction, what about the best performers in terms of either criteria? Sen reports that the rank correlation coefficient is weak and insignificant in the best performers by either criteria. Thus toward the lower as well as the upper end of the distribution of countries 340 Comment Table 2. Slowest Reducers of Under-Fiuc Mortality, 1960-85 Percentage of PubiNc population expend itum with aci to Per capita income on health as baaltth sernices, (doWars) percentage of GNP Cournhry 198S-87 1976 198S 1960 1985 1. Afghanistan 29 160 - - - 2. Somalia 27 110 180 0.6 0.2 3. Ethiopia 46 100 110 0.7 1.0 4. Nigeria 40 380 800 0.3 0.4 5. Rwanda 27 110 280 0.5 0.6 6. South Africa - 1,340 2,060 0.5 0.6 7. Uganda 61 240 - 0.7 0.2 8. Bangladesh 45 110 150 - 0.6 9. Central African Rep. 45 230 260 1.3 1.2 10. Pakistan S 170 380 0.3 0.2 11.- Sierra Leone - 200 350 - 0.7 12. Swaziland _- - - 13. Gambia, The - - - - - 14. Iran 78 1,930 3,502^ 1.1 2.6 15. Senegal 40 390 370 1.5 1.1 16. Burundi 61 120 230 0.8 0.7 17. Guinea 32 150 320 1.0 .1.0 18. Mozambique 39 170 160 - 1.8 19. Mali 15 100 1S0 1.0 0.7 20. Malawi 80 140 170 0.2 2.4 - Not available. Note: The average share of the population with access to health services for all countries for 1985-87 is 61 percent. a. United Nations 1987 (tables 14 and 19). Sources.: World Bank 1977, 1987; UNDP 1990. (twenty countries at each end of the distribution by either GDP growth rate or mortality reduction rate), there was no significant relation between the rate of income growth (decline), on the one hand, and the percentage reduction in under-five mortality, on the other. For the entire group of 118 countries, how- ever, there was a significant rank correlation. If we omit the worst and best performers (twenty countries at either end), there seems to be a threshold at which the correlation between the rankings of countries by the two criteria becomes significant. It is also important to examine whether Sen's results hold for different sub- periods. For example, when the exercise was repeated for 1980-90, which has been called a decade of lost opportunities for most developing countries, the results were the same as those for the whole period (that is, 1960-85).1 Income declined in more countries, and more countries fell below one-half the mean rate of reduction in mortalty (see tables 3 and 4). islam 341 Table 3. Countries With Slowest Growth of GWIP Per Capita, 1980-90 (percent) Per capita Annual raid rndugion Country growtrh of NP in under-flue mortality 1. Libya -9.20 2.90 2. United Arab Emirates -7.20 3.60 3. Trinidad and Tobago -6.00 4.20 4. Saudi Arabia -S.60 3.60 S. Niger -4.S0 1.60 6. Mozambique -4.10 -1.00 7. Jordan -3.90 4.40 8. C6ted'lvoire -3.70 2.10 9. Nigeria -3.00 1.70 10. Zambia -2.90 1.80 11. Bolivia -2.61 2.60 12. Gabon -2.60 1.70 13. Haiti -2.30 2.30 14. Madagascar -2.30 2.00 15. Rwanda -2.20 1.50 16. Kuwait -2.20 S.80 17. Venezuela -2.20 1.S0 18. Panama -2.20 3.30 19. Syria -2.10 4.00 20. Guatemala -2.10 3.20 21. Peru -2.00 2.20 22. Somalia -1.80 1.40 23. Mauritania -1.80 l.S0 24. Argentina -1.80 2.90 25. Philippines -1.S0 2.30 26. Sierra Leone -1.S0 l.S0 Sources: uNIcEF 1992,1993. Convergence or Otherwise Sen refers to a lack of convergence in mortality rates among countries over a period of time. While it is true that countries with high mortality rates did not achieve higher rates of reduction than the low-mortality countries (in fact, the reverse was true), an analysis of disparity over time in absolute levels of mortal- ity presents a different and more complicated picture. The absolute disparity in terms of mortality rates between countries with "very high" mortality and those with "middle" mortality (as defined by UNICEF) widened from 109 per thousand in 1960 to 153 per thousand in 1980 and 161 per thousand in 1991.2 Similarly, the disparity also widened between the "very high" group and the "high" group. By contrast, the disparity between countries categorized as having "middle" and "low" mortality narrowed significantly over the years, falling from 126 per thousand in 1960 to 52 per thousand in 1980 and 25 per thousand in 1991 (see table 5). 342 Comment Table 4. Slowest Reducers of Under-Five Mortality, 1980-90 (percent) Annual Annual rate of Differnwe growth rate redwction of under-fwi from of GDP Country mortality half-mean percapita 1. Angola -1.10 -3.6 8.10 2. Mozambique -1.00 -3.5 -8.20 3. Norway 0.00 -2.5 6.80 4. Romania 0.60 -1.9 -5.30 5. Afghanistan 0.90 -1.6 0.00 6. Lebanon 1.00 -1.5 0.00 7. Denmark 1.10 -1.4 7.20 8. Uganda 1.30 -1.2 -16.10 9. Mali 1.30 -1.2 4.30 10. Somalia 1.40 -1.1 -9.10 11. BurkinaFaso 1.50 -1.0 0.00 12. Rwanda 1S0 -1.0 3.50 13. Venezuela 1.50 -1.0 -5.00 14. Sierra Leone 1.30 -1.0 -2.30 15. Guinea 1.50 -1.0 4.30 16. Paraguay 1.50 -1.0 -0.60 17. Mauritania 1.50 -1.0 3.80 18. Bangladesh 1.60 -0.9 5.40 19. Chad 1.60 -0.9 5.10 20. Niger 1.60 -0.9 -1.20 21. Bhuman 1.60 -0.9 0.00 22. Burundi 1.60 -0.9 0.40 23. Guinea-Bissan 1.60 -0.9 0.00 24. Nepal 1.60 -0.9 1.80 25. Cameroon 1.70 -0.8 2.80 26. Ghana 1.70 -0.8 -10.80 Soumes: uNiCEr 1992,1993. If percentage differences rather than absolute differences are considered, the conclusions are broadly unchanged. The percentage difference in mortality rates between "very high" and "high," between "high" and "medium," and between "very high" and "low" mortality countries increased over time. The percentage difference between "middle" and "low" mortality countries was altmost unchanged between 1960 and 1980 but then declined between 1980 and 1991. This general trend persists if one examines the percentage change in the reduc- tion of mortality rates between these two periods. An acceleration in the reduc- tion of mortality rates for countries in both "high" and "middle" categories (from 1.6 to 3.1 percent for the first group and from 4.4 to 5.8 percent for the second) paralleled a slowdown in reduction rates for countries in the "very high" and "low" categories (from 1.1 to 1.0 percent for the first group and from 5.0 to 4.1 percent for the second). In other words, the "middle" and "high" mortality countries caught up with the "low" mortality group. Islam 343 Table 5. Absolute Disparity in Levels of Under-Five Mortality (per 1,000 live births) Under-five martality (tbousands) Item 1960 1980 1991 By level of under-five mortality Very high 283 222 197 High 231 165 116 (52) (57) (81) Middle 174 69 36 (57) (96) (80) Low 48 17 1 1 (126) (52) (25) Least developed countries' 286 222 186 (69) (84) (85) All developing countries 217 138 101 Industrial countries 45 23 17 By region Sub-Saharan Africa 261 203 180 Middle East and North Africa 246 145 90 (15) (58) (90) South Asia 238 179 131 (23) (24) (49) East Asia and Pacific 198 80 42 (63) (123) (138) Latin America and the Caribbean 161 89 57 (100) (114) (123) Note: Very high (above :40); high (70-140); middle (20-70); low (below 20). The figures in parentheses in the first part of the table indicate the differece in the levels of under-five mortality foraed group and the group immediately preceding it for each of the three different years. lbe figures in parentheses in the second part of the table indicate the differecce between each group and Sub-Saharan Africa. a. The poorest devdoping countres, as defined by the United Nations on the basis of income per capita, pacentage of GNP derived from industry, literacy rate, and other fctors. So0e: UNICEF 1993, table 10. Choice of Under-Five Mortality Sen has written extensively on the subject of choosing social indicators other than GNP for measuring progress, arguing that well-being has to do with being well-that is, living a long, healthy, and literate life. His use of under-five mortality for measuring regress in this paper is not explained, but one can advance several reasons for it. First, it measures an end result of the develop- ment process rather than an input or a means to an end, such as school enroll- ment, per capita calorie availabilitv, or ratio of doctors to population. Second, the under-five mortality rate is in the nature of a composite index; it captures the combined results of a wide variety of inputs and incorporates the effect of such variables as the mother's nutritional status, access to health services, income and food availability, dean water and transportation, and- so on. Third, it is less susceptible to the fallacy of the average: the natural scale does not allow children 344 Comment of the rich to be a thousand times as likely to survive, even if the man-made scale does permit them to have a thousand times as much income.3 Therefore it represents a more accurate picture of society as a whole. And last, data on under-five mortality rates are available for many countries over a long period, making it possible to undertake both cross-sectional and time-series analysis.4 Sen emphasizes the relationship between income growth and improvement in social indicators of well-being. He distinguishes between growth-led and support-led strategies for social improvement. The former relies on economic growth to expand private income-especially of the poor-and raise public expenditure on social services, health, education, and nutrition. The latter undertakes an expansion of social expenditure without waiting for an increase in income and resources generated by higher growth. Its success depends on the "discriminating" use of national resources, the efficiency of public services, and a redistributive bias in delivery. Sen agrees that economic growth is essential for sustaining a support strategy in the long run. Balancing the two strategies may pose a dilemma; both make a daim on financial and administrative resources. Achieving a low level of under-five mortality in relation to income through large or extensive public spending on social services may divert resources that would otherwise be used for investment. If, as a result, income growth slows, absolute mortality may end up higher than it would have been with less public support and more growdt (Drtze and Sen 1989). Life Expectancy as a Focal Variable Using life expectancy as a focal variable (see Anand and Ravallion 1993) pro- duces broadly similar conclusions; there is no correlation between the worst performers in income growth and the worst performers in improved life expec- tancy (tables 6 and 7). No country in this latter category recorded a decline in income. In fact, income growth in 4 of 23 countries in this category was higher than the median income growth for the 118 countries in Sen's list, and in only 3 countries was growth less than 1 percent of GDP per capita. Moreover, all of the worst performers in income growth improved life expectancy at a higher rate than that achieved by the worst performers in terms of the life expectancy criteria. Thus it appears that the dissonance between two perspectives, one based on per capita income growth and the other on life expectancy, is in fact much greater than the dissonanc. between under-five mortality and income growth. What is more intriguing is that there is a dissonance between infant mortality and life expectancy. The divergence between these (apparently) related indica- tors of social improvement is surprising and requires further analysis. Explaining Economic Decline in Africa In whatever sense one defines it, economic regress in Africa has been much discussed and there has been an oversupply of explanations. Recently, atten- Islam 345 Table 6. Slowest Improvers of Life Expectancy and GDP Per Capita, 1960-85 (pcent) Annual increase int Annual increase Economy life expectancy in per capita GDP 1. Paraguay 0.14 2.56 2. Uruguay 0.25 0.04 3. Belize 0.29 4. Argentina 0.30 0.60 S. Trinidad and Tobago 0.33 1.98 6. Cyprus 0.41 4.69 7. Laos 0.43 8. Mauritius 0.44 2.36 9. South Africa 0.46 1.61 10. Sri Lanka 0.49 1.39 11. Singapore 0.50 5.96 12. Rwanda 0.51 1.37 13. Uganda 0.52 2.S7 14. Hong Kong 0.5S 6.18 15. Colombia 0.56 2.30 16. Bhutan O.S6 17. Barbados 0.57 2.97 18. Cambodia 0.58 19. Burundi 0.60 0.73 20. Jamaica 0.61 1.05 21. Suriname 0.61 2.74 22. Mexico 0.65 2.51 23. Venezuela 0.65 1.14 - Not available Sources: UNrDP 1990, 1992; Sen, in this volume, table 3. tion has focused on the collapse of "governance." What is considered of para- mount importance is transparent, accountable, honest, and competent gc' ernment-a requirement not easy to meet in most countries. Experience shows that a government can effectively implement only a few things at a time. Policymakers in Africa-with assistance from the international community- need to concentrate on a limited number of high-priority tasks in the next five years or so. One important aspect of economic regress in Africa that has not been ade- quately emphasized is the movement and growth of vast refugee populations fleeing wars and civil strife. The resulting drain on resources, compounded by ethnic tensions and political instability, has undoubtedly contributed to eco- nomic decline throughout the region. A second factor is that, as the table below shows, military expenditure as a share of GNP has increased over the past three decades much faster in Sub- Saharan Africa than in developing countries as a whole (UNiXp 1991). 346 Comment Military Public expenditure expenditure as an haldh as Per capita GDP percentage of GNP percentage of GNP (dollars) Region 1960 1986 1960 198S 196U 1988 Sub-Saharan Africa 0.7 3.3 0.7 0.9 640 1,180 All developing countries 4.2 5.5 1.0 1.4 790 2,170 Second, while the disparity in per capita income between the two groups has widened, the disparity in terms of military expenditure has rapidly narrowed. Finally, in 1960 Africa's military expenditures and its public expenditures on health accounted for an equal percentage of GNP. By 1980 military spending was four times higher. Sen argues that developing countries like those in Africa can significantly expand social expenditures on health with modest resources and at the same time favor the poor by shifting priorities toward health services in rural areas and focusing on preventive care, rather than, say, channeling support to urban hospitals.5 There are, however, complicating factors. In some countries in Africa wage structures for technical skills or for urban health workers do not always reflect market forces. Salaries are out of line with income levels and budgetary capabilities. As a result, a given amount of public or private expenditure does not buy the amount of services that uMlcEFs low- wage hypothesis suggess (table 8). South Asia spends half as much as Africa on health to serve the same propor- dion of population. With a slightly higher percentage of expenditure, developing Table 7. Slowest-Growth-Rate Countnies and Life Expectancy, 1960-8S (pc-=cl Annal Increas in cauntry GDP per capita ife expectaney 1. Kuwait -5.82 0.77 2. Mozambique -1.95 0.93 3. Angola -1.75 1.16 4. Madagascar -1.66 0.99 5. Chad -1.42 1.03 6. Zambia -1.17 0.90 7. Ghana -0.91 0.66 8. Zaire -038 0.85 9. Central African Rep. -036 0.97 10. Guyana -0.44 0.89 11. Liberia -0.39 0.78 12. Mali -0.29 1.12 13. Afghanistan -0.27 0.92 14. Nicaragua -0.09 0.91 15. Somalia -0.05 0.99 16. Benin -0.02 1.36 Sowrw uXP 1990,1992; Sen, this volume, table 3. islam 347 Tabic 8. Access to Health Services and Government Expenditure on Health Percentage of Percentage of population with access centralggovernment to health services, gevpenditures on health, Region 1985-91 1986-91 Sub-Saharan Africa 52 4 Northern Africa and Middle East 77 6 South Asia 52 2 East Asia and Pacific 87 Latin America and Caribbean 72 6 Least developed countries 45 5 All developing countries 75 5 - Not available. Source: UNICEF 1993. countries as a whole provide health services to 50 percent more people. Similar concdusions regarding the high cost of health services in Africa are shown in the table below (UNDP 1990, 1991). Public heaith Population Population expenditures as per doctor, per nurs, Nurses percentge of GNP, 1984 1984 perdoctor, Region 1986 (tbousands) (thousands) 1984 Sub-Saharan Africa 0.8 24.6 2.2 12.4 AU developing countries 1.4 4.8 1.9 2.8 As a percentage of GNP, Africa spends only 40 percent less than all developing countries as a group, but the ratio of population to doctors is a fifth that in all developing countries. Condusion Given the dissonance between different variables-or indicators of regress-as well as between individual social indicators, I would ask, first, whether we need a composite index of focal variables. If this is not desirable or feasible, how are we to choose between the different indicators? What are the poli4c implications of divergent movements over time in different indicators? Second, Sen's analysis uses percentage change in one or the other focal vari- ables over a period of twenty-five to thirty years. But do we not need to recog- nize situations in which a period of decline follows a period of progress in terms of any of the indicators? The 1980s and 1990s were such periods. For policy purposes, this is an important and relevant question. Third, the need for an appropriate balance between investing in productive activities and investing in human capital remains as important as ever in view of the increasing stringency of resources. To suggest that investment in human capital stimulates income growth and therefore that there is no tradeoff or 348 Comment conflict between the two types of expenditure is too rimplistic. In the short run there is a tradeoff. Moreover, the demand for social expenditure is and will be beyond the level at which it is directly related to growth in income or produc- tivity; it is desired as a direct component of well-being. What kind of practical guidance can analysis provide to policymakers in this respect? Fourth, what accounts for the fact that the percentage reduction in under-five mortality or improvement in life expectancy continues unabated, in spite of not only a slowdown in income growth but also an absolute decline in per capita income? What are the implications in the context of policy reforms and struc- tural adjustment programs that result in economic stagnation or declining income in the short or medium term? Fifth, under-five mortality is significantly negatively correlated with access to health services. Moreover, the rank correlation coefficient between public expenditure on health and mortality rates is, as expected, significantly negative, and public expenditure on health is positively related to access to health services. However, when these relationships are tested separately for two groups of coun- tries that are included in the upper and lower ends of the distribution, they are significant in terms of the relationship between mortality and access to health services, but not significant for the other two relationships. These tentative findings, as well as the circumstances that underlie them, need to be tested with additional data so that we can choose policy instruments to reduce under-five mortality rates. Notes 1. Income growth rates for 196540 and 1980-90 are World Bank figures as quoted in u'ccF (1992, 1993) and are not stricdy comparable with data on income growth in Sen's paper. 2. Very high mortality is defined as 140 per thousand live births, high mortality as 71-140 per thousand, middle mortality as 21-70 per thousand, and low mortality as 20 or less per thousand. 3. The choice of the average annual rate of reduction in undler-five mortality rate for measuring progress reflects the fact that limits are reached with increasing difficulty. At lower levels a given absolute reduction represents a higher percentage reduction. A 10 percent reduction represem a higher rate of progress at lower levds than at higher levels. A fall of 10 points from 100 w 90 represents a reduction of 10 percent, whereas a 10-point faU from 20 to 10 represents a 50 percent reducdon (UwCEF 1993, p. 83). 4. In his earlier writings Sen recognized thatalthough infantand child mortality was by no means to be interpreted as a summary index of the quality of life as a whole, or even of the nutritional saus of the population, it was a usefl indicator (Dreze and Sen 1989, p. 184). One wonders why Sen did not use the Human Development Index, popularized by uwp since 1990, which uses a composite index of literacy, life expecan, and per capita income. However, consistent data over'a number of years may not be available for evaluating performance of countries over time. S. Depending on the level of infrastructure and if the physical facilities and trained personnel are already in placc (these are big "ifs"), it should be possible to provide essential primary care at a recurrent cost of $10 per capita a year. 'For even the poorer countnres with GNP per capita around S200, the public expenditure for gradually achieving universal primary health care would be no more than 2 percent of GNP, increasing to 3 percent by 2000, provided measures for improving efficiency are adopted" (World Bank 1989). Islam 349 References Anand, Sudhir, and Martin Ravallion. 1993. "Human Development in Poor Countries: On the Role of Private Incomes and Public Services." journal of Economic Perspectives 7 (Winter): 133-50. Dr*ze, Jean, and Amartya K. Sen. 1939. Hungerand Public Action. Oxford, U.K.: Clarendon Press. Summers, Robert, and Alan Heston. 1991. "The Penn World Table (Mark 5): An Expanded Set of International Cornparisons, 1955-1988." Quarterly oumnal ofEconomics 106 (May): 327-68. uNDP (United Nations Development Programme). 1990. Human Development Report. New York: Oxford University Press. - . 1991. Human Development Report.New York: Oxford University Press. UNICEF (United Nations Children's Fund). 1987. The State of the World's Children. New York: Oxford University Press. .1992. The State of the World's Childre. New York: Oxford University Press. -. 1993. The State of the World's Children. Ncw York: Oxford University Press. United Nations. 1937. Statistical Yearbook. New York. World Bank. 1987. World Development Report 1987. New Yorc: Oxford University Press. -. 1989. Sub Sahara Africa: From Crisis to Sustanable Growth. Washington, D.C. .1992. World Development Report 1992. New Yurk. Oxford Univcrsity Press. 350 Comment FLOOR DIscussiON OF THE SEN PAPER R esponding to discussant Nora Lustig, Sen denied reporting that there was no relationship between gross domestic product (GDP) per capita and a reduction in under-five mortality; what he had said was that the relationship was not especially dose. Among the top ten performers in reduction of under-five mortality, five were high-growth countries; among the bottom ten performers, three had negative GDP growth. So there is a relationship, and it would not be right to conclude that GDP plays no role in mortality reduction. But much depends on how cDp is-used and on other influences, as well. Health and education are labor-intensive operations that can be affected by GDP growth in two ways: (a) increased GDP would make more resources available for health and education, but (b) as wages go up, health and education become more expensive. These two factors do not cancel each other out, and on the whole, a country is better off with higher GDP. Despite the net gain, however, not all the increase in GDP can be translated into an equivalent amount of reduced infant mortality. Still, Sen took Lustig's point, and he also agreed with the comments by Islam (discussant). There was a good reason for not using the human development index, Sen commented. After all, GNP is a component of that index, so -he would have observed a stronger relationship, but he wanted to use something different and better. He could have used education, but it seemed to him that there was a great gain in separating out the influence of different factors. For the analysis, one wants to look at a detailed pattern, but for reporting results, a narrow focus is better. If one says seventeen things, people don't pay attention;; they are more likely to pay attention if there is one sharply focused contrast with GNP. Both Lustig and Islam had observed that the infant mortality rate can continue to decline even when GNP performance is unsatisfactory. That, said Sen, is because other influences operate, induding the extension of medical knowledge and the delivery of public health services. In Bangladesh, for example, the use of This session was chaired by Robert Picciotto, director-general, Operations Evaluation, the World Bank Provceedngs of the World Bank Annuad Conference on DLevelopment Economics 1993 0 1994 The Intemational Bank for Reconstruction and DeVelopment I m RE tsA 3w51 oral rehydration therapy radically changed the pattern of deaths from diarrhea and cholera. There was some evidence of a continued reduction in the mortality rate in Bangladesh at a time when other indicators were not improving. One participant observed that for smaller groups, such as the top and bottom income groups in the population, the same relationships are not seen as in the large group. That is often true statistically, said Sen. The charitable explanation is that there is a threshold effect and the difference has to be very big to have an impact; in smaller groups the differences are smaller, and in a longer series random and measurement variable errors will even out. The uncharitable expla- nation is that when one is looking at big changes, everything tends to correlate with everything else. The answer is not that one should look only at the long series. Rather, one should not be too discouraged at not finding a fine-tuned difference among groups of similarly placed countries. A participant asked Sen if he thought the asymmetry between progress and regress had anytiing to do with the difference between a country's having stock (physical capital, labor, and institutions of knowledge) that greatly affects eco- nomic progress and having knowledge (for example, -of oral rehydration ther- apy, which reduces infant mortality even if the economy is doing badly). Sen agreed that knowledge contributed to the asymmetry. Knowledge has a way of lingering, he said, so even when a country regresses economically, medical Inowledge does not disappear immediately. Eventually it might,l but perhaps not for a long time. To expand medical knowledge, however, investments of resources might be needed even though GDP is shrinking. The physical capital for production, as well as medical knowledge and epidemiological infrastructure, are factors in both economic growth and the reduction of infant mortality. And the asymmetry is not as strong as one might think. Five of the ten winners in terms of reducing under-five mortality were high-growth countries, and three of the losers were low-growth countries. ,Another participant proposed a classification of countries into warfaring states and trading states. Warfaring states would be those focused on preempt- ing power and grabbing- territory; trading states would be those focused on commerce, the pursuit of wealth, and enrichment. Presumably the goal is to transform -warfaring-state mentalities into trading-state mentalities. How can this be achieved when most of the- policy reforms advocated test the limits of social and political cohesion in the countries in which they are implemented? Sen was grateful for a chance to comment on this issue as it applied to Sub- Saharan Africa. We pay too little attention, he said, to how much Africa suf- fered from the way the cold war was conducted. African countries-being smaller than counties in Asia and Latin America-were already prone to ten- sion and division. During the cold war the United States did not worry about whether a country happened to be undemocratic or would not tolerate opposi- tion, as long as it supported the United States. Similarly, the U.S.S.R did not worry about a country's being nonegalitarian and nonsocialistic, as long as it was on the Soviet side. As a result, between the 1950s and the 1980s Africa 352 FloorDicussion changed dramatically. In 1963, when Sen was first in Africa, he gave a lecture in Uganda and was impressed with the quality of African journalism. He wrote a piece for the Indian paper Economic Weekly commenting on how much more politically probing African journalists were than the docile Indian journalists. By the end of the decade, not only was that no longer true, but 90 percent of the people he had met in Africa had disappeared. Repression had been ushered in, with the connivance of both sides in the cold war. Understanding what has happened in Sub-Saharan Africa, said Sen, requires not only observing that countries, even regions, arcn't large enough units; it requires recognizing how deeply the global cold war had harmed the continent. A World Bank participant asked if Sen thought that the issue of under-five mortality might not mask the marked deterioration in what people eat. How could one unravel the complexities between mortality and illnesses that are linked to hunger and malnutrition, at least in childhood (there being few indices for adults)? Sen agreed that hunger may increase even as the mortality rate declines and said that hunger is an equally important indicator of the quality of life. How one interprets these indicators is a complex question, however. It came as a surprise to him, when he was examining hunger and nutrition in some West Bengal villages, how often undemutrition was connected with parasitic diseases, so that in some ways the reduction of disease had an impact on the level of nutrition. One does not get the full picture of nutrition by monitoring only food intake; the food needs to be marshalled for bodily functions. Nutritional charaaeristics, as well as the amount of food consumed, must be taken into account. Another participant from the World Bank asked Sen if he had -considered examiing the correlation between declining mortality and other economic vari- ables, such as outward-looking strategies. Was there a strong association between an outward orientation and declining mortality? Sen responded that this was a complex question. An outward orientation does have an impact on the reduction of under-five mortality, especially in East Asia, but only when it is associated with growth of GDr. But about half of the slow-groi ing-countries that were winners in terms of reduced infant mortality-including China, Costa Rica, and Jamaica-were not really outward-looking in that period. The impact of outward orientation comes mainly from its impact on economic growth, or on GDP. Of course, oral rehydration therapy is also part of an outward orienta- tion in health services because it was transmitted from one country to another. A participant from the European Investment Bank wondered whether oral rehydration therapy might not be swamping many other effects and influencing the mortality reduction variable-whether it might not be an example- of what Paul Romer, at a previous conference, had called an idea gap. Sen agreed that the under-five mortality rate might record changes that look big but are in fact not because other factors may be moving more slowly. Under-five mortality is intrinsically- valuable to consider on its own, added--Sen, but it is interesting to look at it together with other indicators. Sen and Dr&ze had found that longevity FloorDicuion 353 and under-five mortality tend to go together, with one exception: in Chile £ under-five mortality had behaved very well, but longevity had not done so well. Lustig agreed that the oral rehydration therapy project was an example of a project that bridged the idea gap and said that one of the most valuable roles an institution like the World Bank could play was to make such ideas available to the world. 354 Floor Discussion