You are on page 1of 8

Social Science & Medicine 72 (2011) 617e624

Contents lists available at ScienceDirect

Social Science & Medicine


journal homepage: www.elsevier.com/locate/socscimed

World health inequality: Convergence, divergence, and development


Rob Clark*
Department of Sociology, University of Oklahoma, Kaufman Hall 331, 780 Van Vleet Oval, Norman, OK 73019, USA

a r t i c l e i n f o
Article history: Available online 16 December 2010 Keywords: Cross-national Development Health inequality Life expectancy Infant mortality

a b s t r a c t
Recent studies characterize the last half of the twentieth century as an era of cross-national health convergence, with some attributing welfare gains in the developing world to economic growth. In this study, I examine the extent to which welfare outcomes have actually converged and the extent to which economic development is responsible for the observed trends. Drawing from estimates covering 195 nations during the 1955e2005 period, I nd that life expectancy averages converged during this time, but that infant mortality rates continuously diverged. I develop a narrative that implicates economic development in these contrasting trends, suggesting that health outcomes follow a welfare Kuznets curve. Among poor countries, economic development improves life expectancy more than it reduces infant mortality, whereas the situation is reversed among wealthier nations. In this way, development has contributed to both convergence in life expectancy and divergence in infant mortality. Drawing from 674 observations across 163 countries during the 1980e2005 period, I nd that the positive effect of GDP PC on life expectancy attenuates at higher levels of development, while the negative effect of GDP PC on infant mortality grows stronger. 2010 Elsevier Ltd. All rights reserved.

Introduction The last half of the twentieth century has been characterized as an era of cross-national health convergence (Neumayer, 2003; Wilson, 2001), as well as an era featuring rapid economic growth for a number of developing nations (Firebaugh, 2003). However, the extent to which these two trends are connected, and whether economic development is responsible for welfare gains in the developing world, are hotly debated topics (Brady, Kaya, & Beckeld, 2007; Firebaugh & Beck, 1994; Wimberley & Bello, 1992). Some research celebrates the centrality of development as a primary mechanism for improving human welfare among less developed countries (Firebaugh & Beck, 1994). From this perspective, economic development is thought to introduce a higher standard of living via better wages (Firebaugh & Beck, 1994), more advanced medical technology (Shen & Williamson, 2001), and the purchase of goods that directly or indirectly improve health (Pritchett & Summers, 1996), thereby reducing mortality. The benets of development are also thought to trickle down, improving welfare via a number of intermediary mechanisms, such as domestic investment and formal education (Jenkins & Scanlan, 2001). Previous studies have found strong associations between a countrys level of economic development (or its rate of growth

* Tel.: 1 405 325 2566. E-mail address: robclark@ou.edu. 0277-9536/$ e see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2010.12.008

over time) and human welfare outcomes. Development has been found to positively affect life expectancy and negatively affect infant/child mortality, even among poorer countries (Babones, 2008; Scanlan, 2010; Wickrama & Mulford, 1996). Economic development has been shown to improve a nations physical quality of life, a composite measure consisting of items such as life expectancy and infant mortality (London & Williams, 1990). Similarly, economic growth has been shown to reduce infant mortality, while boosting life expectancy, among developing nations (Firebaugh & Beck, 1994; Shandra, Shandra, & London, 2010). In sum, previous studies generally conclude that a nations level of economic development and/or its growth rate are positively associated with improvements in human welfare, particularly among less developed countries. However, other work suggests that economic growth may not have played a large role in mortality improvements during the twentieth century (Preston, 1975). Scholars have drawn attention, anecdotally, to notable discrepancies between a countrys level of health and its level of development (Sen, 1999; Shen & Williamson, 2001). Other research shows that economic growth produces only modest welfare benets in the developing world (Wimberley & Bello, 1992). Indeed, even when economic development has been found to signicantly improve a countrys infant survival rate (Shen & Williamson, 2001), child survival rate (Shen & Williamson, 1997), or life expectancy (Brady et al., 2007) among developing countries, it oftentimes produces effects that are smaller than other predictors, such as education or gender-related measures.

618

R. Clark / Social Science & Medicine 72 (2011) 617e624

Amartya Sen and others have concluded that a single-minded pursuit of economic development is far too narrow and restrictive. It is as important to recognize the crucial role of wealth in determining living conditions and the quality of life as it is to understand the qualied and contingent nature of this relationship.Without ignoring the importance of economic growth, we must look well beyond it (Sen, 1999: 14). In particular, it is important to consider the distribution of income and wealth within nations (Stiglitz, Sen, & Fitoussi, 2010). While economic growth may suggest that a countrys living standards have risen, the welfare of vulnerable segments of the population may not improve when the distribution of benets is highly skewed. For example, as Jenkins and Scanlan (2001: 719) note, economic development may increase a countrys aggregate food supply, but the poor and disadvantaged remain hungry when resources are unequally controlled or sectoral disparities (urban vs. rural; industrial vs. agricultural) favor some segments of the population over others. If inequality increases enough relative to the increase in average per capita GDP, most people can be worse off even though average income is increasing (Stiglitz et al., 2010: 3). Thus, inequality may reduce the health benets of economic development by distorting the gains achieved from growth. In sum, while a number of previous studies have been able to trace welfare improvements to economic development, others suggest that this relationship is weak or conditional across welfare outcomes. Consequently, it is uncertain whether the crossnational health convergence observed by some during the twentieth century can be attributed to economic development. In this study, I offer new evidence regarding (a) inequality trends in life expectancy and infant mortality, as well as (b) the role that economic development may have played in producing these trends. I rst examine whether life expectancy averages and infant mortality rates have indeed converged across 195 countries during the 1955e2005 period. Consistent with prior work, I nd that cross-national inequality in life expectancy declined during the 1955e2005 period, but that this convergence stalls during the post1990 era. Moreover, and contrary to previous work, I nd that infant mortality rates diverged continuously across the sample period. I then develop a narrative to explain these contrasting trends, suggesting that cross-national health outcomes follow a welfare Kuznets curve. According to this model, economic development improves life expectancy more than it reduces infant mortality among poor countries, whereas the situation is reversed among wealthier nations. To test this thesis, I use random effects models to estimate the impact of gross domestic product per capita (GDP PC) on life expectancy and infant mortality, drawing from 674 observations across 163 countries during the 1980e2005 period. In these models, GDP PC signicantly raises life expectancy and reduces infant mortality, producing massive effects that are larger than any other substantive predictor under investigation. However, the results also show that GDP PCs positive effect on life expectancy weakens at higher levels of development (thereby contributing to cross-national convergence), while its negative effect on infant mortality becomes stronger (thereby contributing to cross-national divergence). In sum, I argue that economic development has contributed to both convergence in life expectancy and divergence in infant mortality. Methods Dependent variables Life expectancy Life expectancy at birth indicates the number of years a newborn infant would live if prevailing patterns of mortality at the time of its birth were to stay the same throughout its life.

Infant mortality Infant mortality refers to the number of infants dying before reaching one year of age, per 1000 live births in a given year. Data for both measures come from the World Population Prospects: The 2006 Revision (United Nations, Department of Economic and Social Affairs, Population Division, 2007). The data are reported in veyear intervals across the 1955e2005 period. Independent variables Unless otherwise noted, all predictors come from the World Development Indicators (World Bank, 2010). I log several variables to reduce skew, as noted below. GDP PC (PPP) (log) Following previous studies (Babones, 2008; Brady et al., 2007; Scanlan, 2010; Shandra et al., 2010; Shen & Williamson, 2001), I measure economic development with each countrys GDP PC based on purchasing power parity. Data are in constant 2005 international dollars. An international dollar has the same purchasing power over GDP as the U.S. dollar has in the United States. In the models presented below, I also include a second-order term for GDP PC to test whether the impact of economic development on life expectancy and infant mortality weakens/strengthens among wealthier nations. Time period In order to control for the substantial rise in life expectancy averages and decline in infant mortality rates, I include a time period indicator in the subsequent analyses (1 1985; 2 1990; 3 1995; 4 2000; 5 2005). World region I also control for the considerable cross-sectional variation in welfare by including world region as a predictor in the analyses. I classify states as belonging to one of the following six world regions: (1) Europe and the West (the excluded reference category), (2) Latin America and the Caribbean, (3) Central and Sub-Saharan Africa, (4) North Africa and the Middle East, (5) East Asia and the Pacic, and (6) Eastern Europe and Central Asia. HIV prevalence Previous work suggests that the AIDS crisis stalled the longterm convergence trend in life expectancy (Goesling & Firebaugh, 2004; Neumayer, 2004) and has had a signicant impact on child mortality among less developed countries (Scanlan, 2010). Accordingly, I include a measure of each countrys HIV prevalence rate, which refers to the percentage of people ages 15e49 who are infected with HIV. Because the AIDS crisis emerged during the 1990s, the World Development Indicators does not report HIV prevalence rates during the early portion of my sample period. Thus, to preserve this period for my analyses, all pre-1990 values for this measure are coded as zero. Following Brady et al. (2007), I also include controls for school enrollment, fertility rate, and urbanization, each of which were found to be signicant predictors in their models of female/male life expectancy and infant/child mortality. School enrollment School enrollment refers to the ratio of secondary school enrollment to the population of the age group that ofcially corresponds to the secondary level. Fertility rate (log) Fertility rate represents the number of children that would be born to a woman if she were to live to the end of her childbearing

R. Clark / Social Science & Medicine 72 (2011) 617e624

619

years and bear children in accordance with current age-specic fertility rates. Urbanization Urbanization refers to the proportion of the total population living in urban areas. In addition, I control for a countrys level of democracy and domestic investment. Democratization Prior studies suggest the importance of democracy for improving a countrys life expectancy and infant mortality (Wickrama & Mulford, 1996), as well as its overall physical quality of life (London & Williams, 1990). Democracy scores come from Freedom in the World 2010 (Freedom House, 2010). Freedom Houses annual survey measures freedom according to two broad categories: political rights and civil liberties. Each country is rated on a sevenpoint scale in both categories, with 1 representing the most free, and 7 representing the least free. I calculated each countrys average score across both categories and inverted this value so that higher numbers represent greater levels of democracy. Capital formation I measure domestic investment with each states level of capital formation, calculated as a share of GDP. Capital formation considers additions to the xed assets of the economy, including land improvements (e.g., fences, ditches, drains), plant, machinery, and equipment purchases, as well as the construction of roads, railways, schools, ofces, hospitals, private residential dwellings, and commercial and industrial buildings. Previous work has found that domestic investment reduces infant mortality and positively affects health spending, the latter two of which are signicantly associated with one another in the developing world (Wimberley, 1990). Sample My sample includes 674 observations across 163 countries over ve waves during the 1980e2005 period. For the dependent variables, the ve waves cover the years 1985, 1990, 1995, 2000, and 2005. For the predictors, each wave represents a prior ve-year period (1980e1984, 1985e1989, 1990e1994, 1995e1999, and 2000e2004), where each data point for each measure represents a states average value across the entire wave. The pooled data are unbalanced with some states contributing more observations than others. However, almost all states (157) are present across two or more waves. In separate diagnostics, I re-estimated models with the full set of predictors for life expectancy and infant mortality when (a) restricting my sample to those 157 countries contributing two or more observations (N 668), and (b) using a balanced sample comprised of those 100 countries that contributed observations across all ve waves (N 500). In both sets of replications, the main effect of GDP PC remains statistically signicant (p < .001) as a positive predictor of life expectancy and negative predictor of infant mortality. Moreover, GDP PCs second-order term continues to reduce the positive effect of life expectancy (p < .05) and increase the negative effect of infant mortality (p < .001). I also considered the impact of inuential observations by removing outliers formally identied in those models featuring the full set of predictors using the Hadi procedure available in Stata 11 (Stata Corporation, 2009). The procedure identies multiple outliers in multivariate data using ordinary least squares (OLS) regression. In the life expectancy model, I detected 13 outliers (Botswana, wave 5; Equatorial Guinea, waves 4e5; Lesotho, waves 3e4; Mongolia, waves 1e2; Rwanda, wave 3; United Arab Emirates, wave 1; Zimbabwe, waves 1e3 and 5). In the infant mortality

model, I detected seven outliers (Equatorial Guinea, waves 4e5; Lesotho, waves 3e4; Mongolia, waves 1e2; United Arab Emirates, wave 1). When re-estimating my models without these cases, the main effect of GDP PC remains statistically signicant (p < .001) as a positive predictor of life expectancy and negative predictor of infant mortality. Moreover, GDP PCs second-order term continues to reduce the positive effect of life expectancy (p < .05) and increase the negative effect of infant mortality (p < .001). Analysis My dataset contains multiple observations for different countries across time. Such a panel structure allows me to use estimation techniques that deal with potential heterogeneity bias (the confounding effect of unmeasured time-invariant variables), which can seriously affect OLS coefcient estimates. The random effects model (REM) and the xed effects model (FEM) represent two estimation strategies designed to correct for the problem of heterogeneity bias. Both procedures simulate the unmeasured time-invariant factors as country-specic intercepts (Nielsen & Alderson, 1995: 685). I present results using a generalized least squares estimator of the REM that includes a rst-order autocorrelation correction. The REM represents the matrix weighted average of the estimates produced by the FEM (focusing on variation within cases) and the between effects model (BEM) (focusing on variation between cases). Thus, while the FEM ignores all cross-sectional variation in its estimates by subtracting all variables from their case-specic means, and the BEM ignores all temporal variation by using the case-specic means as predictors, the REM incorporates both the xed and between effects components, thereby making the REM advantageous for capturing both cross-sectional and longitudinal variation. In short, REMs are useful because it is important to examine how economic development predicts variation in welfare outcomes across countries, as well as how economic growth predicts variation in welfare outcomes within individual countries over time. Moreover, because much of the variation in life expectancy and infant mortality is cross-sectional (see Table 2), FEMs would crucially ignore much of the variation that exists in the data. And because the number of states (N 163) far exceeds the number of waves (T 5), it is important to make efcient use of the crosssectional variation. In fact, it has been shown that FEMs produce biased results when N far exceeds T (Nickell, 1981), as is the case in the present study. FEMs also consume a degree of freedom for every additional N, making them less efcient than REMs (Greene, 2000). Results Inequality trends In the rst stage of analysis, I examine inequality trends in life expectancy and infant mortality based on data from 195 nations during the 1955e2005 period. Previously, Neumayer (2004) reports inequality trends in life expectancy and infant mortality using the standard deviation, keeping both health measures in their original form. As he notes, keeping measures in their original form can be problematic when examining mean-changing variables, as the standard deviation will rise (if the mean rises) or fall (if the mean falls) even if the distributional pattern remains constant. According to Neumayer (2004), there are a few remedies to this problem, including (1) logging the variables prior to calculating the standard deviation so that the results are not sensitive to changes in the mean, or (2) dividing the standard deviation by the mean (i.e., calculating the coefcient of variation). However, Neumayer (2004: 733) dismisses these remedies, noting that it makes little difference for the analysis of convergence trends whether variables are

620 Table 1 Cross-national health inequality (1955e2005), N 195. Year SDOR Mean SDLG .239 .231 .220 .207 .195 .191 .174 .166 .179 .175 .183 .587 .653 .699 .744 .786 .852 .914 .960 1.016 1.063 1.101

R. Clark / Social Science & Medicine 72 (2011) 617e624

COV .232 .223 .210 .196 .183 .176 .161 .153 .157 .158 .165 .468 .519 .561 .601 .644 .722 .755 .809 .874 .903 .946

Gini .133 .128 .121 .112 .104 .099 .091 .085 .086 .087 .091 .267 .297 .321 .342 .365 .398 .419 .443 .470 .485 .502

Theil .027 .025 .023 .020 .017 .016 .014 .012 .013 .013 .015 .117 .144 .167 .189 .214 .256 .284 .319 .360 .384 .415

SDOR Life expectancy 12.547 12.069 11.142 9.947 9.322 8.791 8.156 7.899 7.977 8.382 8.902

Mean

SDLG

COV .255 .233 .205 .172 .156 .142 .130 .123 .122 .126 .132 .459 .496 .507 .546 .584 .610 .655 .685 .701 .727 .763

Gini .138 .127 .113 .097 .088 .080 .072 .068 .066 .068 .071 .255 .274 .282 .308 .330 .341 .366 .380 .384 .393 .407

Theil .031 .026 .021 .015 .012 .010 .009 .008 .008 .008 .009 .124 .145 .149 .164 .183 .197 .225 .241 .247 .260 .280

Life expectancy (population-unweighted) 1955 11.746 50.572 1960 11.811 53.059 1965 11.614 55.243 1970 11.194 57.222 1975 10.830 59.101 1980 10.688 60.896 1985 10.114 62.758 1990 9.870 64.361 1995 10.268 65.276 2000 10.486 66.168 2005 11.082 66.985 Infant mortality (population-unweighted) 1955 60.018 128.360 1960 59.246 114.230 1965 57.014 101.673 1970 54.305 90.336 1975 51.828 80.507 1980 51.556 71.425 1985 46.508 61.576 1990 44.186 54.623 1995 43.421 49.665 2000 40.599 44.940 2005 38.542 40.759

(population-weighted) 49.183 .246 51.785 .227 54.314 .202 57.786 .177 59.867 .162 61.698 .150 62.866 .135 64.387 .129 65.428 .132 66.516 .137 67.441 .144

Infant mortality (population-weighted) 62.904 136.944 .657 61.021 122.993 .706 51.820 102.286 .707 47.382 86.806 .725 44.653 76.418 .754 41.719 68.396 .796 38.933 59.480 .852 36.349 53.100 .880 34.300 48.958 .894 32.814 45.136 .920 31.522 41.326 .945

Note: SDOR Standard deviation when the variables are kept in their original form; SDLG Standard deviation when the variables are logged; COV Coefcient of variation.

held in log or in level form. Hence any potential bias is too small to affect the analysis here, and that the coefcient of variation is approximately equal to the standard deviation of the natural logarithm of the variable. In Table 1, I examine the matter empirically, reporting inequality trends in life expectancy and infant mortality using (1) the standard deviation when the variables are kept in their original form (SDOR), (2) the standard deviation when the variables are logged (SDLG), (3) the coefcient of variation (COV), as well as two popular index measures, (4) the Gini, and (5) the Theil, both of which are scaleinvariant (i.e., inequality is not affected by the mean). The top panel shows the life expectancy trends (both population-unweighted and population-weighted). In this case, even though the mean rises somewhat across time (31%e37%, depending on whether countries are population-weighted), it matters little which measure is used. According to all ve measures, and regardless of whether I weight countries based on their population size, inequality in life expectancy generally declines from 1955 to 1990 and increases from 1990 to 2005. This is consistent with prior work noting the overall convergence in life expectancy, but featuring a reversal at the end of the period due to the AIDS crisis in Africa (Goesling & Firebaugh, 2004; Neumayer, 2004). However, the inequality trends in infant mortality (shown in the bottom panel) reveal quite different results. According to SDOR, inequality declines between 1955 and 2005. However, the mean drops dramatically, as well (68%e70%, depending on whether countries are population-weighted), suggesting that this may account for the drop in SDOR. Not surprisingly, when adopting either remedy to the changing mean (SDLG or COV), or when relying on the Gini or the Theil, cross-national inequality in infant mortality increases steadily throughout the sample period. Moreover, these results hold whether I weight countries by their population size or not. In sum, the overall trend for life expectancy is one of convergence (notwithstanding the post-1990 reversal), while the inequality trend in infant mortality is quite the opposite, diverging steadily across the sample period. In order to illustrate how these contrasting trends have manifested, I summarize life expectancy averages and infant mortality

rates across six world regions in Table 2. The rst two columns in Table 2 show the 1955 and 2005 means, respectively, for life expectancy. In 1955, life expectancy was highest in the West (68.18), followed by Eastern Europe (60.53). As the third column indicates, the average life expectancy increased in every region of the world during the sample period. However, the growth rate in life expectancy was lowest in the two regions where life expectancy was highest in 1955: the West (16.04%) and Eastern Europe (17.69%). Every other world region increased their life expectancy by at least twice as much, with the Middle East (56.49%) and East Asia (50.83%) leading the way. Thus, life expectancy averages converged because the healthiest countries (i.e., the West and Eastern Europe) experienced the least improvement. However, the nal three columns tell a very different story for infant mortality trends. In 1955, infant mortality was lowest in the West (39.74), while it was highest in Africa (180.03). As the last column indicates, infant mortality rates dropped dramatically in every region of the world between 1955 and 2005. However, the reduction in infant mortality rates was greatest in the West (86.02%) and lowest in Africa (50.32%), while all other world regions reduced infant mortality by more or less the same rate (77.02% to 79.14%). Thus, infant mortality rates diverged during the sample period because the healthiest (i.e., the West) and unhealthiest (i.e., Africa) regions continued to diverge from one another. The Welfare Kuznets Curve In this section, I develop a narrative to help explain the contrasting trends in cross-national health inequality that have occurred during the last half of the twentieth century: convergence in life expectancy averages and divergence in infant mortality rates. In doing so, I suggest that economic development may have played a role in producing both trends, capitalizing on the idea from prior research that the impact of economic development does not tend to be uniform across outcomes or samples. As Brady et al. (2007: 5) suggest, GDP PC might not have consistent effects across all measures of well-being. In a prior health study, Pritchett and Summers (1996: 845) divide their cross-national sample into four

R. Clark / Social Science & Medicine 72 (2011) 617e624 Table 2 Regional trends and patterns, life expectancy and infant mortality (1955e2005), N 195. Life expectancy Mean (1955) All states (N 195) Europe & the West (N 23) Latin America & the Caribbean (N 37) Central & Sub-Saharan Africa (N 47) North Africa & the Middle East (N 25) East Asia & the Pacic (N 34) Eastern Europe & Central Asia (N 29) 50.57 68.18 53.84 38.93 45.14 46.70 60.53 Mean (2005) 66.98 78.97 72.41 51.72 69.43 68.92 70.93 Growth rate 35.77% 16.04% 36.42% 34.15% 56.49% 50.83% 17.69% Infant mortality Mean (1955) 128.36 39.74 108.71 180.03 168.32 133.80 99.13 Mean (2005) 40.76 4.84 22.68 90.43 37.28 33.35 23.50

621

Growth rate 72.44% 86.02% 78.79% 50.32% 79.14% 77.02% 78.27%

Note: Europe & the West Australia, Austria, Belgium, Canada, Channel Islands, Denmark, Finland, France, Germany, Iceland, Ireland, Italy, Luxembourg, Malta, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States; Latin America & the Caribbean Argentina, Aruba, Bahamas, Barbados, Belize, Bolivia, Brazil, Chile, Colombia, Costa Rica, Cuba, Dominican Republic, Ecuador, El Salvador, French Guiana, Grenada, Guadeloupe, Guatemala, Guyana, Haiti, Honduras, Jamaica, Martinique, Mexico, Netherlands Antilles, Nicaragua, Panama, Paraguay, Peru, Puerto Rico, Saint Lucia, Saint Vincent-Grenadines, Suriname, Trinidad-Tobago, Uruguay, Venezuela, Virgin Islands (USA); Central & Sub-Saharan Africa Angola, Benin, Botswana, Burkina Faso, Burundi, Cameroon, Cape Verde, Central African Republic, Chad, Comoros, Congo (DR), Congo (R), Djibouti, Equatorial Guinea, Eritrea, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea-Bissau, Ivory Coast, Kenya, Lesotho, Liberia, Madagascar, Malawi, Mali, Mauritania, Mauritius, Mozambique, Namibia, Niger, Nigeria, Reunion, Rwanda, Sao Tome-Principe, Senegal, Sierra Leone, Somalia, South Africa, Swaziland, Tanzania, Togo, Uganda, Zambia, Zimbabwe; North Africa & the Middle East Afghanistan, Algeria, Bahrain, Cyprus, Egypt, Iran, Iraq, Israel, Jordan, Kuwait, Lebanon, Libya, Morocco, Oman, Pakistan, Palestine, Qatar, Saudi Arabia, Sudan, Syria, Tunisia, Turkey, United Arab Emirates, Western Sahara, Yemen; East Asia & the Pacic Bangladesh, Bhutan, Brunei, Cambodia, China, East Timor, Fiji, French Polynesia, Guam, Hong Kong, India, Indonesia, Japan, Laos, Macao, Malaysia, Maldives, Micronesia, Mongolia, Myanmar, Nepal, New Caledonia, North Korea, Papua New Guinea, Philippines, Samoa, Singapore, Solomon Islands, South Korea, Sri Lanka, Thailand, Tonga, Vanuatu, Vietnam; Eastern Europe & Central Asia Albania, Armenia, Azerbaijan, Belarus, Bosnia-Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Georgia, Greece, Hungary, Kazakhstan, Kyrgyzstan, Latvia, Lithuania, Macedonia, Moldova, Montenegro, Poland, Romania, Russia, Serbia, Slovakia, Slovenia, Tajikistan, Turkmenistan, Ukraine, Uzbekistan.

income quartiles. And while they do not draw much attention to it, their results show that the wealthiest two quartiles increased their life expectancy by lower rates, but reduced their infant mortality by higher rates, than the poorest two quartiles during the 1960e1990 period. Thus, poor countries had been outperforming rich countries in improving life expectancy, but lagging behind in their reduction of infant mortality. This suggests, rst, that indicators of economic development may not perform optimally across all welfare measures in the developing world. Indeed, using a sample of less developed countries during the 1980e2003 period, Brady et al. (2007) nd that GDP PC positively affects male and female life expectancy, but that it does not signicantly impact infant or child survival. In short, the above ndings suggest that the link between development and life expectancy may be stronger among poorer nations, whereas the link between development and infant mortality may be stronger among wealthier nations. To pursue this idea further, consider the classic Kuznets curve, a phenomenon in which income inequality is lowest within poor and wealthy nations and greatest within middle-income nations. As poor nations begin to develop and industrialize, inequality within these nations begins to rise for two reasons: (1) industrial wages are higher than agricultural wages, and (2) there is greater economic inequality within the industrial sector than the agricultural sector (Kuznets, 1955). Thus, the initial shift to industrial production is accompanied by rising inequality, as a small (but growing) number of people begin entering into a higher-paying sector whose wages show greater dispersion around the mean. It is only until advanced industrialization that income inequality peaks and begins to dissipate, where agricultural workers constitute a small (and shrinking) portion of the workforce, thereby corresponding to the falling portion of the Kuznets curve. Although the pattern is typically detected in a cross-section of countries, the curve is presumably operative within individual countries over time, as well. In the same vein, consider what I call a welfare Kuznets curve that models the complex relationship between economic development and health outcomes. As the urban/industrial transition commences, the bulk of the population (which disproportionately includes the impoverished) remains excluded. Improvements in life expectancy occur more quickly than reductions in infant mortality, given that infant and child mortality rates are more accurate

welfare gauges of the broader population (and of the poor, in particular) (Brady et al., 2007: 5) and are less sensitive to welfare gains made disproportionately by the wealthy (Brady et al., 2007: 10). As countries continue to develop and join other nations that reside closer to the middle of the income distribution, the disparity between life expectancy (which is steadily improving) and infant mortality (which is improving at a slower pace) reaches its peak. By contrast, during the downward phase of the curve, as the agricultural/rural segment of the population begins to constitute a relatively small (and dwindling) percentage of the total, further increases in development begin to yield diminished returns for life expectancy. On the other hand, improvements in infant/child mortality start to accelerate, as the broader population becomes increasingly situated within a context of elevated living standards. Consequently, the welfare Kuznets curve offers an explanation for both (a) the long-term convergence in life expectancy, as most of these improvements occur early in development, and (b) the longterm divergence in infant mortality, as most of these reductions occur later in development. Is this model consistent with the observed data? In order to illustrate the welfare Kuznets curve empirically, I rst created a set of world health residuals by regressing life expectancy on infant survival (i.e., the inverse of infant mortality). I then plotted the residuals against a nations GDP PC among all 674 cases in my sample. Fig. 1 shows the resulting scatter plot along with a tted quadratic line. While there is considerable scatter, the observations tend to cluster in a curvilinear fashion. GDP PC is positively associated with the residuals among the poorest countries (r .495 before the inection point of the curve), but negatively associated with the residuals among wealthier nations (r .657 after the inection point of the curve). That is, economic development improves life expectancy more than infant survival among poor nations, whereas the situation reverses among wealthier nations. One way to appreciate the practical signicance of this curvilinear association is to divide the 674 observations in my sample into one of two groups based on their location relative to the inection point in the curve (GDP PC $2530) and compare their welfare trends during the 1980e2005 period. Table 3 reports the results of this comparison. Group A features 243 observations (comprising 36% of the sample) located before the curve (GDP PC < $2530), with a large majority of these observations coming

622

R. Clark / Social Science & Medicine 72 (2011) 617e624

World Health Residuals

-1

-2

-1

GDP PC (PPP) (Log) (Standardized)

Fig. 1. The Welfare Kuznets Curve (1980e2005), N 673a. aOne observation (Rwanda, wave 3) not pictured in order to preserve scale. Note: World Health Residuals produced by regressing life expectancy on infant survival. Positive residuals indicate a higher life expectancy than expected given that countrys infant survival rate. Negative residuals indicate a lower life expectancy than expected given that countrys infant survival rate.

from Africa (62.14%) and East Asia (22.63%). Group B features 431 observations (comprising 64% of the sample) located after the curve (GDP PC > $2530), with a majority of these observations coming from Latin America (26.22%) and the West (25.06%). The overall comparison featured in the rst row indicates that states belonging to observations in Group A improved their life expectancy by 12.84% during the 1980e2005 period, while states belonging to observations in Group B only improved their life expectancy by 10.14%. However, countries in Group A only reduced their infant mortality by 34.84%, while those in Group B cut their infant mortality by 58.79% (note that some countries may have crossed the inection point during the sample period and would, therefore, be contributing observations to both groups). Thus, countries found before the inection point increased their life expectancy to a greater extent, but decreased their infant mortality to a lesser extent, than countries found after the inection point. Moreover, as

the subsequent rows indicate, this pattern holds even when comparing countries within the same world region. Consequently, in the following analyses, I hypothesize that economic development will signicantly improve a countrys life expectancy average and infant mortality rate. However, I anticipate that GDP PCs positive effect on life expectancy is weaker at higher levels of development, while its negative effect on infant mortality is stronger at higher levels of development. Analyses Table 4 presents results from random effects models of life expectancy (models 1e4) and infant mortality (models 5e8). Each cell reports the z-score, with the standardized coefcient in bold. In model 1, GDP PC is a positive predictor of life expectancy (p < .001), while its second-order term exerts a negative effect (p < .01), indicating that the impact of GDP PC on life expectancy becomes signicantly weaker at higher levels of development. These two measures by themselves explain over 70% of the variation in life expectancy (R2 Overall .713). In model 2, I add the temporal and regional controls. As expected, time period is positively signed and highly signicant (p < .001), indicating the longitudinal increase in life expectancy across the sample period. Also, all of the regional measures are negatively signed and signicant, indicating that all ve regions exhibit lower life expectancy averages than the West. Nevertheless, the main effect of GDP PC (B .41; p < .001) and its second-order term (B .06; p < .01) remain highly signicant. In model 3, I add the substantive controls, all of which are signicant at the .05 level or higher, and in the expected directions. Again, though, GDP PC (B .21; p < .001) and its second-order term (B .05; p < .01) remain signicant. Moreover, the main effect of GDP PC exerts the largest impact among the substantive predictors, followed by fertility rate (B .18; p < .001), urbanization (B .12; p < .001), HIV prevalence (B .11; p < .001), and school enrollment (B .10; p < .001). Democratization (B .05; p < .01) and capital formation (B .02; p < .05) also achieve signicance as positive predictors, but their effects are smaller. In sum, GDP PC

Table 3 Change in life expectancy and infant mortality (1980e2005), N 674. Life expectancy (growth rate) Group A All states Europe & the West Latin America & the Caribbean Central & Sub-Saharan Africa North Africa & the Middle East East Asia & the Pacic Eastern Europe & Central Asia 12.84% (N 243) e (N 0) 13.65% (N 11) 7.51% (N 151) 20.75% (N 15) 26.95% (N 55) 3.87% (N 11) Group B 10.14% (N 431) 7.62% (N 108) 12.55% (N 113) 2.90% (N 36) 18.69% (N 75) 13.47% (N 47) 3.85% (N 52) Infant mortality (growth rate) Group A 34.84% (N 243) e (N 0) 48.01% (N 11) 25.38% (N 151) 41.76% (N 15) 55.38% (N 55) 39.42% (N 11) Group B 58.79% (N 431) 61.52% (N 108) 57.00% (N 113) 29.25% (N 36) 69.81% (N 75) 62.31% (N 47) 58.35% (N 52)

Notes: Group A consists of 243 observations located before the inection point (GDP PC < $2530) of the curve featured in Fig. 1. Group B consists of 431 observations located after the inection point (GDP PC > $2530) of the curve featured in Fig. 1.

R. Clark / Social Science & Medicine 72 (2011) 617e624 Table 4 Random effects models of life expectancy and infant mortality (1980e2005) Life expectancy Model 1 GDP PC GDP PC2 Time period Latin America Africa Middle East East Asia Eastern Europe HIV prevalence School enrollment Fertility rate Urbanization Democratization Capital formation Gini index GDP PC Gini index Observations States R2 Within R2 Between R2 Overall 674 163 .139 .734 .713 674 163 .294 .837 .816 674 163 .660 .863 .863 22.06d .67 2.99c L.07 Model 2 11.77d .41 2.73c L.06 9.16d .22 2.78c L.12 10.01d L.51 4.58d L.16 3.47d L.15 2.14b L.09 Model 3 5.93d .21 2.78c L.05 6.30d .16 0.23 L.01 4.53d .22 0.84 L.03 0.70 L.03 2.75c L.09 18.18d L.11 3.75d .10 5.53d L.18 3.36d .12 3.09c .05 2.33b .02 Model 4 3.78d .20 Infant mortality Model 5 26.79d .86 4.56d .11 Model 6 15.17d L.49 2.51b L.05 22.86d L.38 7.76d .36 12.09d .64 9.73d .38 6.72d .32 4.11d .18 Model 7 12.33d L.39 5.30d L.09 9.18d L.20 4.32d .16 5.68d .27 5.04d .17 2.80c .11 4.17d .14 8.08d .04 4.64d L.10 10.32d .30 1.53 L.05 2.59c L.03 0.20 .00

623

Model 8 9.46d L.50

1.46 .06 0.88 .04 2.94c L.19 0.50 .02 0.42 L.02 1.80a L.07 10.66d L.09 4.09d .17 3.60d L.19 0.98 .04 2.14b .05 3.33d .06 1.42 .03 0.85 L.02 287 128 .646 .899 .884

674 163 .359 .775 .772

674 163 .727 .844 .852

674 163 .799 .906 .907

3.97d L.15 2.24b .11 2.13b .14 3.28c .15 1.34 .07 3.93d .15 3.39d .03 3.36d L.13 6.45d .31 0.06 .00 1.79a L.04 0.10 .00 2.65c .06 2.76c .06 287 128 .815 .924 .890

Notes: All models include a rst-order autocorrelation correction. Each cell reports the z-score with the standardized coefcient in bold. a p < .1 b p < .05 c p < .01 d p < .001(two-tailed tests)

exerts large positive effects on a countrys life expectancy average, but it more effectively improves life expectancy among poorer nations than it does among wealthier nations. In models 5e7, I turn to infant mortality. In model 5, both the main effect (p < .001) and the second-order term (p < .001) for GDP PC are highly signicant as negative predictors of infant mortality, indicating that the negative effect of GDP is strengthened substantially among wealthier nations. Impressively, these two measures by themselves explain over 77% of the variation in life expectancy (R2 Overall .772). In model 6, I add the temporal and regional controls. Again, as expected, time period is negatively signed and highly signicant (p < .001), indicating the longitudinal decline in infant mortality across the sample period. Meanwhile, all of the regional measures are positively signed and reach the highest level of signicance (p < .001), indicating that all ve regions exhibit higher infant mortality rates than the West. Nevertheless, the addition of these controls does not explain away the curvilinear effect of GDP PC, as both the main effect (B .49; p < .001) and second-order term (B .05; p < .05) remain signicant. In model 7, I introduce the substantive predictors. Net of these measures, GDP PCs main effect (B .39; p < .001) and second-order term (B .09; p < .001) remain signicant. In fact, GDP PC again exerts the strongest effect among all the substantive predictors, followed

by fertility rate (B .30; p < .001) and school enrollment (B .10; p < .001). Collectively, the ndings from Table 4 are consistent with the thesis that economic development had a major impact on human welfare over the past several decades, but that its differential performance across samples and health outcomes has produced contrasting trends, helping life expectancy averages to converge and infant mortality rates to diverge. Finally, I examine the impact of income inequality on the relationship between economic development and health. Theoretically, inequality may exert a harmful moderating effect by concentrating the health benets associated with economic development into the hands of the privileged classes. However, because infant mortality is a more accurate welfare gauge of the poor than life expectancy, inequality should reduce the negative association between development and infant mortality more than it reduces the positive association between development and life expectancy. I examine this proposition in models 4 and 8, where I estimate the direct and moderating impact of income inequality using a reduced sample of countries for which inequality estimates are available. I use the Gini index to measure inequality (World Bank, 2010), which indicates the extent to which the distribution of income within a country deviates from a perfectly equal distribution. Thus, higher scores indicate greater inequality. I also include an interaction term (GDP

624

R. Clark / Social Science & Medicine 72 (2011) 617e624

PC Gini Index) in these models to test whether the benets of economic development are signicantly reduced in those countries with greater income inequality. In model 4, the interaction term is negatively signed, but not signicant, indicating that income inequality does not signicantly reduce the impact of development on life expectancy. In model 8, however, the Gini index is highly signicant as a positive predictor of infant mortality (B .06; p < .01). Moreover, the interaction term indicates that inequality signicantly reduces the negative effect of development on infant mortality (B .06; p < .01). This latter nding provides support for development critics who claim that it is important to take into account the distribution of benets within a country when considering the impact of development on health.

realm must be made cautiously. The relationship between economic development and welfare is complex. The promotion of growth policies must rst consider which welfare outcomes are being targeted, as well as where such policies are to be implemented. While economic development offers clear benets in most contexts, there may be circumstances in which preventing the spread of HIV or reducing fertility or income inequality should take priority. Likewise, future studies examining the relationship between economic development and human welfare should carefully consider both setting and health outcome. References
Babones, S. (2008). Income inequality and population health: correlation and causality. Social Science and Medicine, 66, 1614e1626. Brady, D., Kaya, Y., & Beckeld, J. (2007). Reassessing the effect of economic growth on well-being in less-developed countries, 1980e2003. Studies in Comparative International Development, 42, 1e35. Firebaugh, G. (2003). The new geography of global income inequality. Harvard University Press. Firebaugh, G., & Beck, F. (1994). Does economic growth benet the masses? Growth, dependence, and welfare in the third world. American Sociological Review, 59, 631e653. Freedom House. (2010). Freedom in the world 2010. Freedom House. Retrieved May 2010. http://www.freedomhouse.org/template.cfm?page439. Goesling, B., & Firebaugh, G. (2004). The trend in international health inequality. Population and Development Review, 30, 131e146. Greene, W. (2000) (4th ed).Econometric analysis, . Prentice Hall. Jenkins, J., & Scanlan, S. (2001). Food security in less developed countries, 1970 to 1990. American Sociological Review, 66, 718e744. Kuznets, S. (1955). Economic growth and income inequality. American Economic Review, 45, 1e28. London, B., & Williams, B. (1990). National politics, international dependency, and basic needs provision: a cross-national analysis. Social Forces, 69, 565e584. Neumayer, E. (2003). Beyond income: convergence in living standards, big time. Structural Change and Economic Dynamics, 14, 275e296. Neumayer, E. (2004). HIV/AIDS and cross-national convergence in life expectancy. Population and Development Review, 30, 727e742. Nickell, S. (1981). Biases in dynamic models with xed effects. Econometrica, 49, 1417e1426. Nielsen, F., & Alderson, A. (1995). Income inequality, development, and dualism: results from an unbalanced cross-national panel. American Sociological Review, 60, 674e701. Preston, S. (1975). The changing relation between mortality and level of economic development. Population Studies, 29, 231e248. Pritchett, L., & Summers, L. (1996). Wealthier is healthier. Journal of Human Resources, 31, 841e868. Scanlan, S. (2010). Gender, development, and HIV/AIDS: implications for child mortality in less industrialized countries. International Journal of Comparative Sociology, 51, 211e232. Sen, A. (1999). Development as freedom. Anchor Books. Shandra, J., Shandra, C., & London, B. (2010). Do non-governmental organizations impact health? A cross-national analysis of infant mortality. International Journal of Comparative Sociology, 51, 137e164. Shen, C., & Williamson, J. (1997). Child mortality, womens status, economic dependency, and state strength: a cross-national study of less developed countries. Social Forces, 76, 667e694. Shen, C., & Williamson, J. (2001). Accounting for cross-national differences in infant mortality decline (1965e1991) among less developed countries: effects of womens status, economic dependency, and state strength. Social Indicators Research, 53, 257e288. Stata Corporation. (2009). Stata 11. Stata Press. Stiglitz, J., Sen, A., & Fitoussi, J. (2010). Mismeasuring our lives: why GDP doesnt add up. New Press. United Nations, Department of Economic and Social Affairs, Population Division. (2007). World population prospects: the 2006 revision. InComprehensive tables, vol. I. United Nations. http://www.un.org/esa/population Retrieved December 2008. Wickrama, K., & Mulford, C. (1996). Political democracy, economic development, disarticulation, and social well-being in developing countries. Sociological Quarterly, 37, 375e390. Wilson, C. (2001). On the scale of global demographic convergence 1950e2000. Population and Development Review, 27, 155e171. Wimberley, D. (1990). Investment dependence and alternative explanations of third world mortality: a cross-national study. American Sociological Review, 55, 75e91. Wimberley, D., & Bello, R. (1992). Effects of foreign investment, exports, and economic growth on third world food consumption. Social Forces, 70, 895e921. World Bank. (2010). World development indicators. World Bank. http://data. worldbank.org/data-catalog Retrieved May 2010.

Discussion Cross-national inequality trends in life expectancy and infant mortality have followed different trajectories over the past half century. Life expectancy averages have demonstrated long-term convergence since the mid-twentieth century, while infant mortality rates have continuously diverged during this same time period. In 1955, Western nations enjoyed the longest life spans and lowest infant mortality rates. Since that time, these countries experienced slower growth in life expectancy, but faster improvement in infant mortality, than other world regions. In this study, I nd that economic development helps account for both crossnational convergence in life expectancy, as well as divergence in infant mortality, via its differential impact on these two welfare outcomes. In generating a narrative to explain this process, I identify a welfare Kuznets curve, in which economic development yields more immediate benets to life expectancy at early stages of growth, with reductions in infant mortality lagging behind. In other words, development improves life expectancy more than it reduces infant mortality among poor nations. However, during the latter stages of development, this dynamic reverses, such that further economic growth yields diminishing returns to life expectancy, while reductions in infant mortality begin to escalate. Drawing from a sample of 674 observations during the 1980e2005 period, I show that countries located before the inection point of the welfare Kuznets curve (GDP < $2530) improved their life expectancy at a higher rate, but reduced their infant mortality at a lower rate, than countries located after the inection point (GDP > $2530). Finally, I estimate the differential impact of economic development on life expectancy and infant mortality. I nd that economic development boosts life expectancy and reduces infant mortality to a greater extent than any other substantive predictor under investigation. However, GDP PCs positive effect on life expectancy is signicantly reduced at higher levels of development, whereas its negative effect on infant mortality is notably enhanced. In general, the ndings from this study support a development agenda. Economic growth is quite important for improving welfare outcomes among both rich and poor societies. Nevertheless, it is important to understand the limitations of development as a welfare tool. It is not a question of whether development improves health, but under which circumstances its effects will be maximized. Moreover, a number of factors may reduce the impact of development. For example, the results from this study indicate that the pursuit of economic development is a less effective strategy for combating infant mortality in those countries featuring high levels of income inequality. More generally, policy prescriptions in this

You might also like