94667 AGGREGATE INCOME SHOCKS AND INFANT MORTALITY IN THE DEVELOPING WORLD Sarah Baird, Jed Friedman, and Norbert Schady* Abstract—Health and income are strongly correlated both within and infant mortality for a large set of developing countries across countries, yet the extent to which improvements in income have a causal effect on health status remains controversial. We investigate between 1975 and 2004. Infant mortality is pervasive in the whether short-term fluctuations in aggregate income affect infant mortal- developing world. In poor countries, approximately 30% of ity using an unusually large data set of 1.7 million births in 59 developing all deaths occur to children under the age of five, compared countries. We show a large, negative association between per capita GDP and infant mortality. Female infant mortality is more sensitive than male to less than 1% in rich countries (see Cutler et al., 2006). infant mortality to negative economic shocks, suggesting that policies that Infant mortality is also much less likely than adult mortality protect the health status of female infants may be especially important to be affected by reverse causality from health to income. during economic downturns. Our focus in this paper is on departures of income from trend, and the effect that these have on infant mortality, I. Introduction rather than on the relationship between long-term changes in income and infant mortality. This is an important distinc- H EALTH and income are strongly correlated across countries and, within countries, across individuals. In the United States, the life expectancy of people in the lowest tion. Even if long-term improvements in infant mortality are caused primarily by improvements in medical technology rather than directly by economic growth, short-term shocks ventile of the income distribution was about 25% lower than to GDP could have important consequences for child health. that of people in the highest ventile in 1980 (Rogot et al., However, the effect of income shocks on infant mortality is 1992). In developing countries, dozens of studies have found hard to sign ex ante. In developing countries, negative that people with higher incomes have better health status and shocks reduce household consumption of nutritious foods lower mortality (see Gwatkin et al., 2007, for a review). The and lower expenditures on other inputs into child health, and seminal work by Preston (1975, 1980) shows that as countries they may seriously disrupt public health services; all of become richer, life expectancy rises, although many other these would tend to increase infant mortality. Aggregate factors are also important in explaining mortality declines. shocks, however, depress wages and imply a lower opportu- Despite the association between income and health status, nity cost of women’s time. Many inputs into the production the extent to which improvements in income have a causal of child health are intensive in parental (especially maternal) effect on health status remains controversial. In an early, time, including taking children for preventive health visits, influential article using cross-country data, Pritchett and breastfeeding, cooking healthy meals, and collecting clean Summers (1996) argued that ‘‘wealthier is healthier,’’ but water. Because systemic shocks reduce the cost of engaging their identification and conclusions have been challenged in these activities, they may improve child outcomes. The by, among others, Jamison, Sandbu, and Wang (2004) and effect of negative income shocks on child health and mortal- Deaton (2006). Part of the concern is the existence of feed- ity is therefore ambiguous in theory. backs from health to income—for example, both Gallup and Since this paper concerns the effect of GDP shocks on Sachs (2001) and the World Health Organization (2001) infant mortality, it is closely related to a literature on the argue that improvements in health status would increase health consequences of booms and busts in aggregate rates of economic growth. Countries with higher income income. Dehejia and Lleras-Muney (2004) conclude that levels also tend to have higher education levels, better-func- infant mortality is generally procyclical in the United States. tioning health systems, and better institutions, all of which A variety of transmission mechanisms have been proposed are likely to improve health outcomes independent of to explain why economic recessions lead to improved child income (Cutler, Deaton, & Lleras-Muney, 2006). health, including reductions in air pollution (Chay & Green- In this paper, we revisit the discussion of the relationship stone, 2003), reductions in health-damaging behaviors such between health and income with an investigation of the as smoking and drinking, and increases in the probability impact of short-term fluctuations in per capita GDP on that mothers engage in time-intensive activities such as exer- cise and prenatal care (Ruhm, 2000; Ruhm & Black, 2002). Received for publication September 12, 2008. Revision accepted for In developing countries, our focus in this paper, the evi- publication January 28, 2010. * Baird: George Washington University; Friedman: World Bank; dence on the relationship between economic downturns and Schady: Inter-American Development Bank. infant mortality is more mixed (see the review by Ferreira & We thank Anne Case, Angus Deaton, Adriana Lleras-Muney, Christo- Schady, 2009). Sharp economic downturns have been asso- pher Mckelvey, David Newhouse, Christina Paxson, Martin Ravallion, three anonymous referees, and seminar participants at the Center for Glo- ciated with increases in infant mortality in Mexico (Cutler bal Development, Princeton University, University of California at Berke- et al., 2002), Peru (Paxson & Schady, 2005), and India (Bha- ley, the Population Association Annual Meetings, and the World Bank for lotra, 2010). Miller and Urdinola (2010), however, find that comments. All remaining errors are our own. The views contained here are solely attributable to us and do not reflect those of the World Bank or arguably exogenous declines in the price of coffee, which the Inter-American Development Bank. resulted in declines in aggregate income in coffee-growing The Review of Economics and Statistics, August 2011, 93(3): 847–856 Ó 2011 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology 848 THE REVIEW OF ECONOMICS AND STATISTICS areas in Colombia, were associated with lower infant mor- tories, closely following Paxson and Schady (2005). Our tality, echoing the results from the United States. measure of infant mortality is an indicator that takes on the Our paper extends the existing literature in a number of value of 1 if a child died at a reported age of 12 months or important ways. The sample, 59 countries, is much larger younger.1 We discard information for children born within than that from the country-specific studies. This allows us twelve months of the survey when calculating mortality to estimate the effect of aggregate income shocks on health rates to avoid complications with censored data. in a variety of settings across recent decades. In addition, Although the DHS are a rich source of data, they also have and unlike the cross-country studies already discussed, we some limitations for our analysis. We briefly discuss two of use individual-level data on infant mortality rather than these limitations, both related to the use of retrospective working with five-year country averages. These data allow information in the DHS to construct birth and death histories. us to control for the changing composition of women giving First, recall bias may be a concern if women are less likely to birth and to assess how aggregate income shocks interact accurately remember more distant births and deaths. To with a variety of characteristics of mothers and children, minimize recall errors, we do not use information on births such as mother’s education and the gender of the child. that occurred more than eleven years prior to the date of the The main finding of this paper is that there is a robust survey. Thus, our birth data cover the period 1975 to 2003.2 relationship between shocks to per capita GDP and infant Second, any given survey is representative of women ages mortality: on average, a 1% decrease in per capita GDP 15 to 49 at the time of the survey, but is not representative of results in an increase in infant mortality of between 0.24 all births and child deaths in earlier years. To see this, note and 0.40 per 1,000 children born. Changes in infant mortal- that a woman aged 49 in a survey carried out in 2000 would ity during economic downturns cannot be explained by the have been 39 in 1990. If no surveys were carried out changing composition of women giving birth. The paper between 1990 and 2000 in the country of study, no data also shows important heterogeneity underlying these aggre- would be available on births to women aged 40 or older in gate results. The mortality of girls is significantly more sen- 1990. Children born to older women may respond to eco- sitive to aggregate economic shocks than that of boys. This nomic fluctuations differently from those born to younger difference is particularly apparent during economic contrac- women. To avoid the problem of varying age composition tions, especially when these are large. This heterogeneity of birth mothers, we discard from the sample births to has important implications for the design of policies to pro- women age 40 or older. Our analysis therefore provides tect children during economic downturns. meaningful estimates of the relationship between income Section II, describes the data for our sample of countries shocks and mortality of infants born to women aged 15 to and provides details on the construction of the variables 39. We note, however, that only 1.2% of births in our sample used in the analysis—in particular, our measure of infant of DHS countries occur to women age 40 or older in any mortality. In section III, we discuss the basic estimation year of analysis. This retrospective construction of births approach and present results. Section IV concludes. and infant deaths to women aged 15 to 39 results in series of varying lengths and with varying start periods depending on II. Data and Construction of Variables the number and dates of DHS surveys in each country.3 The data on per capita GDP used for this paper are taken 1 We use this measure of infant mortality, rather than the standard defi- from World Bank (2007). The values correspond to real per nition of mortality for children younger than 12 months, because of age capita GDP in 2000 U.S. dollars, adjusted for differences heaping in reports of mortality. 2 The results reported in this paper are very similar when a five-year across countries in purchasing power parity (PPP). The data recall is used instead of the ten-year recall. When the recall period is fif- on births and deaths are based on 123 demographic and teen years rather than ten years, our estimates of the impact of GDP health surveys (DHS) covering 59 countries in Africa (33 shocks on infant mortality fall by about one-third, but are still precisely estimated. These results are available from the authors on request. countries, 68 surveys), Latin America (12 countries, 31 sur- 3 In the working paper version of this paper (Baird, Friedman, & veys), and Asia (14 countries, 27 surveys). The earliest sur- Schady, 2007), we show that our estimates of aggregate infant mortality veys in our sample were carried out in 1986, the latest ones are internally consistent and are highly correlated with other sources of data that have been used to assess the relationship between per capita in 2004. Taken together, the data we use contain informa- GDP and infant mortality—in particular, data from the World Develop- tion on approximately 760,000 women and 1.7 million ment Indicators (WDI) database (World Bank, 2007). However, the esti- births. However, the sample sizes vary considerably; for mate of infant mortality we calculate is more useful to estimate the rela- tionship between shocks to per capita GDP and infant mortality for a example, the 1999 India DHS covers approximately 90,000 variety of reasons. First, we have constructed annual series of infant mor- births, while the sample size for the 1987 DHS for Trinidad tality to look at higher-frequency changes than what can be observed in and Tobago is just over 3,800 births. The list of specific sur- the five-year averages in the WDI series. Data like those in WDI will have smoothed some of the year-on-year variation in infant mortality in the veys and their sample sizes is given in table 1. DHS. A share of the variation that is smoothed is likely to be measure- The DHS ask women a set of questions about the date of ment error, but the remainder likely reflects genuine annual fluctuations birth, current vital statistics, and date of death (if deceased) in infant mortality. Second, the data in WDI would not allow us to adjust for the changing composition of women giving birth during economic of all children ever born. We use the responses to these expansions or contractions or to estimate the heterogeneity of responses questions to construct retrospective birth and death his- to economic fluctuations by characteristics of the mother and child. AGGREGATE INCOME SHOCKS AND INFANT MORTALITY IN THE DEVELOPING WORLD 849 TABLE 1.—DHS DATA SETS USED IN THE ANALYSIS Total Number Total Number Country Survey Years of Mothers of Births Armenia 2000 2,446 4,234 Bangladesh 1994, 1997, 1999, 2004 26,313 51,071 Benin 1996, 2001 7,515 18,891 Bolivia 1989, 1994, 1998, 2004 24,574 54,474 Brazil 1986, 1992, 1996 11,672 23,590 Burkina Faso 1993, 1999, 2003 16,362 39,410 Burundi 1987 2,416 6,464 Central African Republic 1995 3,373 7,962 Cameroon 1991, 1998, 2004 11,444 27,350 Chad 1997 4,655 11,829 Colombia 1986, 1990, 1995, 2000 17,149 31,010 Comoros 1996 1,405 3,838 Cote d’Ivoire 1994, 1999 6,660 15,993 Dominican Republic 1986, 1991, 1996, 1999, 2002 23,486 48,458 Ecuador 1987 2,536 5,817 Egypt 1993, 1996, 2000, 2003 33,988 73,378 Ethiopia 2000 8,436 20,484 Gabon 2001 3,371 7,084 Ghana 1988, 1994, 1999, 2003 11,841 25,675 Guatemala 1987, 1995, 1999 13,496 33,832 Guinea 1999 4,549 11,224 Haiti 1995, 2000 7,764 18,283 India 1993, 1999 103,669 208,690 Indonesia 1987, 1991, 1994, 1997, 2003 81,673 153,661 Kazakhstan 1995, 1999 3,971 6,624 Kenya 1989, 1993, 1998, 2003 18,457 44,289 Kyrgyz Republic 1997 2,131 4,100 Liberia 1986 3,419 8,669 Madagascar 1992, 1997 7,592 19,195 Malawi 1992, 2000 11,368 27,292 Mali 1987, 1996, 2001 17,915 47,710 Mexico 1987 4,528 10,177 Morocco 1987, 1992, 2004 14,775 33,052 Mozambique 1997 5,535 12,468 Namibia 1992, 2000 6,674 13,550 Nepal 1996, 2001 12,058 27,569 Nicaragua 1998, 2001 14,098 29,598 Niger 1992, 1998 9,468 26,714 Nigeria 1990, 1999, 2003 14,333 36,543 Pakistan 1991 4,874 13,255 Paraguay 1990 3,208 7,752 Peru 1986, 1992, 1996, 2000 40,330 84,225 Philippines 1993, 1998, 2003 20,621 46,551 Rwanda 1992, 2000 9,317 23,607 Senegal 1986, 1993, 1997 11,881 30,636 South Africa 1998 6,017 9,970 Sri Lanka 1987 4,121 8,250 Sudan 1990 4,242 11,314 Tanzania 1992, 1996, 1999 12,826 29,743 Thailand 1987 4,294 7,516 Togo 1988, 1998 7,611 18,582 Trinidad and Tobago 1987 1,786 3,588 Tunisia 1988 3,224 8,318 Turkey 1993, 1998 7,897 15,306 Uganda 1989, 1995, 2001 11,883 30,062 Uzbekistan 1996 2,315 4,744 Vietnam 1997, 2002 7,643 13,012 Zambia 1992, 1997, 2002 13,776 32,044 Zimbabwe 1989, 1994, 1999 9,346 19,913 Total 764,327 1,668,640 The DHS collect a great deal of current information on health services—for example, prenatal check-ups and the mothers (for example, their education levels, whether they place of delivery. However, these data are not collected in are employed) and children (for example, the gender and a comparable fashion in every survey and typically are birth order and, in the most recent surveys, height and available only for the most recent birth. The degree to weight). Some DHS also ask respondents about their use of which we can analyze possible transmission mechanisms 850 THE REVIEW OF ECONOMICS AND STATISTICS from income to infant mortality with our data is therefore of the survey, the gender of the child, and whether the child limited. was a multiple birth. All of these variables are highly corre- lated with the probability of child survival.5 This approach III. Econometric Specification and Results implicitly assumes that place of residence at the time of the survey is correlated with place of residence at the time of A. Basic Results child birth and that schooling has been completed by age 15; these should be reasonable approximations for most of To estimate the effect of per capita GDP on infant mor- the countries and years in our sample. In addition, as an tality in our data, we pool all surveys and run regressions of alternative means of controlling for compositional effects, the following form: we include a set of mother fixed effects, as well as birth- Dimct ¼ ac þ b log GDPct þ fc ðtÞ þ eimct ; ð1Þ specific characteristics (child gender and an indicator for multiple births). This approach has the advantage that it where Dimct is an indicator variable that takes the value of 1 controls for all time-invariant mother characteristics, not if child i born to mother m in country c in year t died in the just education and place of residence, but limits the sample first year of life, and 0 otherwise; ac is a set of country fixed to women who have had at least two live births. effects; logGDPct is the natural logarithm of per capita Our main set of results is presented in table 2. The first GDP; fc(t) is a flexible, country-specific formulation of time row in table 2, which reports the results from estimates of (in practice, we present results that include linear, quadratic, equation (1), implies that a 1% decrease in per capita GDP and cubic terms); and eimct is the error term. Standard errors is associated with a 0.24 to 0.40 increase in infant mortality are clustered at the country level in order to correct for auto- per 1,000 children born.6 On average, the country-specific correlation of arbitrary form in shocks to infant mortality year-on-year decrease in infant mortality in our data is 2.5 across years within a country. In this specification, b is the per 1,000 live births. A 1% shortfall in per capita GDP from impact of GDP on infant mortality, after removing country- expected trends therefore results in an increase in infant specific trends and intercepts from the data.4 mortality of between 10% and 15% of the average annual In principle, two mechanisms could account for a nega- mortality decline in our data. Note also that those regres- tive association between infant mortality and aggregate eco- sions that more flexibly account for underlying secular nomic circumstances. First, it is possible that a child born trends result in larger (in absolute value) estimates of the to a woman of given characteristics is more likely to die if association between per capita GDP and infant mortality. economic circumstances are unfavorable. Second, it is pos- Previous studies have generally adjusted only for linear sible that the composition of women giving birth changes trends (as in Jamison et al., 2004) and hence may underesti- with economic circumstances. Clearly, these two causes for mate the effect of economic shocks on infant mortality. a countercyclical relationship between GDP and infant mor- Results from regressions that include the vector of cov- tality—changes in mortality risk for a child born to a given ariates Xim, equation (2), are presented in the second row of woman or changes in the pool of women giving birth—have table 2. These results show that including these covariates very different implications for our understanding and for has a negligible effect on estimates of the association the design of corrective policy. between log per capita GDP and infant mortality. The third A direct way to adjust for compositional changes is to row of the table reports the results from regressions that do include the characteristics of women, children, and births in not include mother characteristics or fixed effects for the equation (1), which gives us 5 Dadj imct ¼ ac þ b log GDPct þ fc ðtÞ þ dXimc þ eimct ; ð2Þ There is an extensive literature on this topic. See, for example, the review papers by Behrman and Deolalikar (1988), Strauss and Thomas (1998), and Schultz (2002). where Ximc is a vector of characteristics of child i born to 6 The results in table 2, as well as in the other tables in the paper, differ mother m. Recall that child births and deaths are calculated slightly from those in the working paper version of our paper (Baird et al., 2007) for two reasons. First, the results in the working paper version clus- on the basis of retrospective questions asked of mothers at tered standard errors at the country-year level; in this version, we use the the time of the survey, which limits the variables that can more conservative approach of clustering standard errors at the country be included in equation (2). In practice, we control for level. Second, the results in the working paper version were based on weighted regressions that used the within-country weights provided in third-order polynomials in mother’s years of education, the DHS documentation. In this version, we report the results from maternal age at the time of birth, and birth order and binary unweighted regressions. We choose to show the unweighted regressions indicators for place of residence (urban or rural) at the time since the DHS weights are constructed to draw inferences that are repre- sentative of a country’s population at the time of survey, but our retrospec- tive birth histories extend to eleven years before year of survey, and it is 4 We obtain very similar results from a two-step process in which we not clear how appropriate the weights provided are to the earlier period. first collapse the data to the level of the country-year cell and then More important, we are not conducting a country-by-country analysis, but account for secular trends in various ways, including regressions in first rather pooling data from all available surveys. The weights provided by differences, with a formal error correction model (ECM), and smoothing the DHS make no adjustment for the fact that the underlying populations the data with standard time-series filters such as the Hodrick-Prescott and across countries are very different. For both of these reasons, regressions Baxter-King filters (see Baird, Friedman, & Schady, 2007, for a presenta- without weights are more transparent and intuitive. We note, however, that tion of these estimates). none of the main messages in the paper are affected by these changes. AGGREGATE INCOME SHOCKS AND INFANT MORTALITY IN THE DEVELOPING WORLD 851 TABLE 2.—INCOME SHOCKS AND INFANT MORTALITY TABLE 3.—GDP SHOCKS AND INFANT MORTALITY, INCLUDING POSSIBLE LEAD AND LAG EFFECTS Dependent variable Linear Quadratic Cubic Independent Variable Linear Quadratic Cubic Unadjusted Infant Mortality rate À23.96 À32.88 À39.81 Lagged, current, and lead GDP [8.11]*** [7.40]*** [9.84]*** Lagged GDP À1.08 À6.66 À5.45 Controlling for mother and birth characteristics [10.93] [11.19] [9.85] Infant Mortality rate À23.46 À30.78 À37.83 GDP À31.26 À36.59 À38.74 [7.73]*** [6.99]*** [9.82]*** [11.59]*** [10.43]*** [11.20]*** Unadjusted, restricted to mothers with multiple births Lead GDP 10.93 6.69 6.19 Infant Mortality rate À26.34 À31.08 À38.25 [7.26] [8.59] [8.48] [9.08]*** [7.59]*** [11.60]*** GDP series reweighted to approximate exposure over course of in utero Mothers’ fixed effects Development and first year Infant Mortality rate À29.46 À32.33 À36.22 In utero 7.5 2.37 3.93 [9.43]*** [8.69]*** [11.45]*** [19.86] [17.58] [14.61] Number of observed births equals 1,634,360 in first two panels and 1,356,738 in bottom two panels. First month À38.71 À40.61 À41.77 Mother and birth characteristics are indicators for rural location, gender of child, and multiple birth, and [20.59]* [20.86]* [18.97]** cubic terms for mothers’ age, years of education, and infant birth order. Robust standard errors are clus- Next 11 months 9.84 3.74 1.54 tered at the country level; there are 59 countries. GDP is measured in the year 2000 international (PPP) dollars. *p < .10; **p < .05, ***p < .01. [10.23] [11.01] [12.73] Robust standard errors are clustered at the country level. There are 1,549,745 observations distributed across 840 country-year cells and 59 countries. GDP is measured in year 2000 international (PPP) dol- sample of women who have had at least two live births. lars. *p < .10, **p < .05, ***p < .01. These results are presented to place the fixed-effects esti- mates in context; they show that the association between GDP and mortality in this smaller sample is very similar to that observed in the full sample of live births. The fourth nomic conditions early in the pregnancy that are most row of table 2 reports the results that include the mother important in determining infant mortality; these conditions fixed effects and birth-specific characteristics. These coeffi- are loaded on lagged GDP for most children, and the coeffi- cients are very similar to those without fixed effects. Table cient on lagged GDP is insignificant. Similarly, it does not 2 makes clear that the changing composition of women can- appear to be that conditions in the later part of a child’s first not account for the bulk of the association between infant year in life substantially affect the probability of survival; mortality and aggregate income that we observe in our data. these conditions are loaded on to lead GDP for most chil- Instead, when there are negative economic shocks, there is dren in our sample, and the coefficient on lead GDP is also an increase in mortality risk for an infant born to a given insignificant. Rather, it appears that economic conditions in mother. those months shortly before and shortly after birth have the biggest effect on the probability that a child survives. We make a further attempt to clarify issues about the B. Timing of Shocks to GDP window of vulnerability that infants face with regard to The discussion so far has focused on the contempora- GDP shocks. Mothers report the year and month of birth of neous relationship between GDP and infant death, without each child, and we assign the fifteenth day of the relevant giving explicit attention to the timing of shocks. As a first month as the birth date for each child. Using these data, we step to clarifying this issue, we include terms in lagged and then construct birth-month specific exposure windows for lead per capita GDP in our basic regression.7 The top panel economic conditions in utero, in the first month of life, and in table 3 shows that the coefficients on both of these terms in the next eleven months. The results from these regres- are small, and are not significant at conventional levels. sions are presented in the lower panel of table 3. The coef- Only the coefficient on current GDP in the top panel of ficients on economic conditions in utero and after the first table 3 is significant. This suggests that it is not the eco- month of life are both small and insignificant. By contrast, the coefficient on per capita GDP in the first month is 7 To see how this speaks to the issue of the effects of shocks to GDP at large, significant, and very close in magnitude to that different times in an infant’s life, it is useful to work out what the coeffi- reported in table 2.8 These results underscore that cients on lagged, current, and lead GDP imply for children born at differ- ent times in the year. For a child born early in the year (say, in January), the coefficient on lagged GDP mainly reflects conditions before concep- tion and in utero, the coefficient on current GDP reflects conditions in the 8 first year of life, and the coefficient on lead GDP reflects conditions in the We also experimented with breakdowns of the in utero period. For second year—beyond the period relevant for the measure of infant mor- example, in a study of the effect of the Chernobyl nuclear disaster, tality. By contrast, for a child born late in the year (say, in December), the Almond, Edlund, and Palme (2009) show that radiation exposure was par- coefficient on lagged GDP reflects conditions before conception, the coef- ticularly damaging during the period between 8 and 25 weeks after con- ficient on current GDP reflects conditions in utero, while the coefficient ception. The emphasis of our paper is on economic conditions rather than on lead GDP reflects the conditions after birth. Finally, for a child born at radiation exposure, but it is conceivable that the period of 8 to 25 weeks the midpoint of the year on June 30 (the average birth point in our data), postconception is one in which health insults more generally are particu- lagged GDP reflects conditions before conception and during the first larly damaging. However, in none of the specifications we ran was the three months in utero, current GDP reflects conditions in the last six coefficient on economic conditions in the period corresponding to 8 to 25 months in utero and the first six months after birth, and lead GDP reflects weeks after conception significant once we controlled for conditions in conditions after the infant is six months of age. the last three months of pregnancy and after birth. 852 THE REVIEW OF ECONOMICS AND STATISTICS economic conditions around birth appear to matter most TABLE 4.—INFANT MORTALITY BY MOTHER, CHILD, AND COUNTRY CHARACTERISTICS for infant survival. Characteristic Estimated Infant Mortality Rate The importance of economic conditions around birth for Child gender Boys Girls infant survival also yields clues about the likely transmis- 88.4 78.5 sion mechanisms from aggregate economic shocks to infant (833,545) (800,814) Mother’s education Less Than Primary Primary or Greater mortality. Low birth weight is considered an important risk 99.5 51.3 factor in predicting neonatal and infant death (see, for (1,093,757) (540,603) example, the review by Lawn, Cousens, & Zupan, 2005). Mother’s location Urban Rural 61.4 94.9 However, the fact that the coefficient on economic condi- (555,742) (1,078,618) tions for much of the in utero period is not significant in table 3 suggests that this is unlikely to be the main reason Mother’s age 15–19 20–34 35–39 105.4 77.0 90.1 for elevated infant mortality during economic downturns.9 (296,461) (1,151,038) (144,052) On the other hand, to have a skilled birth attendant during Birth order First Second to Fourth Fifth or More birth, or access to health care for children who face health 79.1 75.6 101.5 381,176 (804,593) (448,591) shocks shortly after, may help explain our findings. Approxi- mately 36% of neonatal deaths worldwide are a result of Country income Low Income Middle Income severe infections during birth, and another 23% are a result 94.5 67.7 (964,446) (669,914) of asphyxia (Lawn et al., 2005). Poor economic conditions Numbers in the sample are in parentheses. around birth could result in either a deterioration in public health services or a decrease in households’ ability to pay for or otherwise access health services related to delivery, as We first present the mean infant mortality rates in table 4 well as pre- and post-natal care (as suggested, for example, for each mother, child, or country characteristic we use in by Paxson & Schady, 2005, in their analysis of infant mortal- our analysis. The first row of the table shows that girls are ity in Peru), both of which could lead to increased mortality eleven percent less likely to die in the first year of life than in the first year of life.10 boys, a well-known finding in the demographic literature.11 The other coefficients show that children born in rural areas are more likely to die than those born in urban areas; that the mortality of children born to mothers with less than pri- C. Heterogeneity mary schooling is almost twice as high as that of children Up to this point, we have implicitly assumed that aggre- born to mothers with completed primary schooling or more; gate income shocks affect all mothers and children equally. that children born to young mothers (ages 15–19) and older However, this need not be so for a host of reasons. For mothers (ages 35–39) are more likely to die than those born example, more educated and wealthier mothers may be bet- to prime-age mothers (ages 20–34); that high-parity births ter able to smooth consumption of critical inputs into child (fifth birth or higher) are also more likely to die than lower- health; there may also be within-household discrimination parity births; and that children born in lower-income devel- so that boys are better protected from negative health oping countries are more likely to die than children born in shocks than girls; families in richer countries may have middle-income countries.12 greater access to credit markets and hence may be better We next analyze heterogeneity in the relationship able to smooth consumption of essential items. We now between detrended per capita GDP and infant mortality turn to the question of heterogeneity of impacts, focusing along these observable dimensions of mothers and children. on differences by the gender of the child, the education Our approach is straightforward. In each case, we generate and age of the mother, place of residence (urban or rural), an indicator for the characteristic in question—for example, birth parity, and the overall income of the country of resi- an indicator for the birth of a girl—and then interact this dence. indicator with the measure of log per capita GDP. Table 5 reports the coefficients on the main effect for log per capita 9 Selection may be important if poor economic conditions in utero lead to a higher rate of spontaneous abortions. The sample of children born 11 alive during bad years may then have higher health endowments, introdu- For example, the World Health Organization (2006) estimates that cing a downward bias to the association we estimate between economic the male-to-female ratio in neonatal mortality and in early neonatal mor- conditions in utero and infant mortality. tality in developing countries is 1.3. 10 12 It is also possible that maternal mortality can play a role in the coun- The GDP data we use in this paper are measured in constant 2000 tercyclical relation between infant survival and GDP. Children whose U.S. dollars. The World Bank (2001) classifies countries as ‘‘low income’’ mothers die in birth are themselves much more likely to die (Anderson if per capita GDP in constant 2000 dollars is below $755. To classify et al., 2007), and maternal death during childbirth may increase during countries as ‘‘low’’ or ‘‘middle’’ income, we apply the World Bank thresh- poor economic times. We thank an anonymous referee for this suggestion. old to the 1980 per capita GDP data. Using 1980, a date before the start of Our birth data are reported retrospectively by mothers alive at time of sur- our infant mortality series for the bulk of the countries we analyze, limits vey, and so we do not observe maternal mortality in our data. As a result, the potential for possible simultaneity biases induced by feedback from if maternal mortality is countercyclical, our estimates would be biased health to income that could arise had we adopted a later date by which to downward. categorize countries. AGGREGATE INCOME SHOCKS AND INFANT MORTALITY IN THE DEVELOPING WORLD 853 TABLE 5.—HETEROGENEITY IN INFANT MORTALITY RATE AND GDP BY MOTHER, 0.53 per 1,000—a remarkable difference by any stan- BIRTH, OR COUNTRY CHARACTERISTIC, CUBIC TREND dard.14 Interaction Characteristic GDP (GDP Â Characteristic) Female infant À27.22 À25.52 D. Magnitude and Sign of Shocks to Per Capita GDP [10.40]** [10.20]** Low mother’s education À31.31 À12.32 In addition to heterogeneity by household characteristics, [11.98]** [14.00] it is possible that the impacts on infant mortality may vary by Rural location À21.28 À26.33 [11.81]* [10.45]** the magnitude of the GDP shock and, perhaps, the direction Young mother (<20 years) À44.20 22.52 (either positive or negative). To investigate this, we estimate [10.15]*** [15.86] Older mother (>34 years) À36.05 À19.37 a series of gender-specific, continuous spline regressions.15 [8.24]*** [18.51] The results from these estimations are reported in table 6, First births À49.24 42.68 separately for boys (upper panel) and girls (lower panel). [10.81]*** [16.26]** High birth order (above fourth) À30.42 À29.78 The top row for each panel presents the results from a spline [9.55]*** [14.54]** regression with a knot at 0, which allows different slopes for Middle-income country À46.14 14.14 positive and negative changes in GDP. We then turn to spline [18.49]** [23.32] regressions with two knots. In the second row, these knots Low mother’s education is defined as less than primary attainment. Robust standard errors are clus- tered at the country level. GDP is measured in year 2000 international (PPP) dollars. In this currency are fixed at À1r and 1r (where r stands for standard devia- measure, the World Bank threshold for middle-income country status is a per capita GNI of $755. *p < .10, **p < .05, ***p < .01. tions of GDP trend deviations in our sample); in the third row, they are fixed at À1.5r and 1.5r; and in the fourth (bottom) row, the knots are fixed at À2r and 2r. Table 6 shows that positive shocks to per capita GDP affect girls and boys in a similar fashion. Negative shocks, GDP and on the interaction between log per capita GDP however, have much larger effects on the mortality of girls and the given characteristic from these regressions. We than boys. For example, for negative shocks of À1.5r or focus on the specification that includes country-specific larger, a 1% decrease in log per capita GDP results in an cubic time trends, as these account for underlying time increase in mortality of 1.05 per 1,000 girls born (with a trends most flexibly. standard error of 0.30), but an increase in mortality of only Table 5 shows that the mortality of infants born to 0.53 per 1,000 boys born (with a standard error of 0.23). On mothers in rural areas is significantly more sensitive to average, countries with a negative shock to per capita GDP changes in economic conditions than that of children born of 1.5r or larger experienced a GDP contraction of 5.9%. to urban mothers. In part, this is the result of the higher (Note that there are 122 such country-year events in our mortality rates among infants born to rural women, data.) The predicted increase in girl infant mortality during although this does not fully explain the differences in the these negative shocks to aggregate income is 7.4 deaths magnitudes we estimate.13 A similar pattern can be seen in per 1,000, approximately three times the magnitude of the a comparison between mothers with less than primary edu- average country-specific annual reduction in mortality. cation and those with primary education or greater—the increase in infant mortality during economic downturns is 14 We also considered gender differences in the impact of shocks to larger for less educated women, but from a higher base— GDP on infant mortality separately by region. These results suggest that the mortality of girls is more sensitive to GDP shocks than that of boys in although this difference is not significant at standard levels. every region in the developing world for which we have data. Thus, in Also, the point estimate on the interaction term for middle- sub-Saharan Africa, a 1% decrease in GDP increases the mortality of boys income countries suggests larger absolute increases in by 0.33 per 1,000, and that of girls by 0.62 per 1,000. Comparable figures for Latin America and the Caribbean are 0.29 per 1,000 (boys) and 0.46 infant mortality during economic downturns in low-income per 1,000 (girls). For Southeast Asia, these are 0.15 per 1,000 (boys) and countries, although this difference is also not significant at 0.24 per 1,000 (girls); for South Asia, 0.72 per 1,000 (boys) and 1.43 per standard levels. The most striking result in the table relates 1,000 (girls); and for the Middle East and North Africa region, a 1% decline in GDP results in a decrease of mortality of 0.18 per 1,000 boys to differences by gender. Although the average mortality born and an increase of mortality of 0.78 per 1,000 girls born. All of the among boys is higher than among girls, table 5 shows that coefficients on GDP in the regressions for girls are significant at the 10% the mortality of girls is much more sensitive to changes in level or higher. In the regressions for boys, only the coefficients in the regressions for Southeast Asia and Latin America and the Caribbean are economic circumstances than that of boys: a 1% change in significant. We conclude that the gender differences in the effect of GDP per capita GDP changes the mortality of boys by approxi- shocks on infant mortality we observe in our sample of developing coun- mately 0.27 per 1,000 children born and that of girls by tries as a whole are not driven by a single region, including regions where a preference for boys has been well documented (for example, South Asia). 15 For this purpose, we regress the indicator for infant death on country fixed effects and a country-specific cubic polynomial in time and predict 13 A 1 log-unit decrease in per capita GDP would increase the infant the residual from this regression. We also regress log per capita GDP on mortality rate of children born to rural women from 95 to 143 and that of country fixed effects and a country-specific cubic polynomial in time and children born to urban women from 61 to 82. The proportional, not just predict the residuals from this regression. We then estimate with a spline the absolute, change among rural mothers is thus larger. regression the relationship between these two residuals. 854 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 6.—HETEROGENEITY IN INFANT MORTALITY RATE AND GDP RELATION BY SIZE OF GDP DEVIATION FROM CUBIC TREND, FOR MALE AND FEMALE CHILDREN Dependent variable Magnitude and Direction of GDP Deviation Infant Mortality Rate for <¼ 0 >0 boys (N ¼ 833,545) À20.75 À31.00 [10.10]** [16.46]* <¼ À1 s.d. > À1 s.d. and <¼ 1 s.d. > 1 s.d. À38.22 À14.73 À40.69 [19.34]* [21.44] [25.10] <¼ À1.5 s.d. > À1.5 s.d. and <¼ 1.5 s.d. > 1.5 s.d. À52.81 À13.16 À67.25 [22.82]** [15.48] [19.67]*** <¼ À2 s.d. > À2 s.d. and <¼ 2 s.d. > 2 s.d. À59.09 À18.07 À70.82 [26.47]** [12.64] [19.65]*** Infant Mortality Rate for <¼ 0 >0 girls (N ¼ 800,814) À55.43 À43.71 [13.90]*** [14.05]*** <¼ À1 s.d. > À1 s.d. and <¼ 1 s.d. > 1 s.d. À75.81 À39.24 À47.27 [24.41]*** [22.00]* [21.84]** <¼ À1.5 s.d. > À1.5 s.d. and <¼ 1.5 s.d. > 1.5 s.d. À104.71 À36.60 À58.66 [30.03]*** [15.97]** [19.93]*** <¼ À2 s.d. > À2 s.d. and <¼ 2 s.d. > 2 s.d. À148.47 À36.52 À69.12 [51.38]*** [14.04]** [19.72]*** Slope coefficients are estimated from a continuous spline specification. Robust standard errors are clustered at the country level. GDP is measured in year 2000 international (PPP) dollars. *p < .10, **p < .05, ***p < .01. These simple back-of-the-envelope calculations suggest IV. Conclusion that the magnitude of the effects of large negative income shocks on infant mortality, in particular of girls, is large by Macroeconomic volatility is a fact for most developing any standard. countries. In recent decades, the standard deviation of In sum, table 6 is consistent either with girls being more income over time has been approximately twice as large in fragile in their first year of life than boys, which seems unli- developing as developed countries (Aguiar & Gopinath, kely, or with families protecting boys more than girls dur- 2007). A recent review stresses the welfare costs of volati- ing economic downturns. In other words, these findings lity for developing countries in terms of their inability to suggest that household behavioral responses to negative smooth consumption (Loayza et al., 2007). In this paper, shocks play an important role in determining infant survi- we document another way in which aggregate economic val. Finally, table 6 underscores that our results are unlikely fluctuations can have dramatic welfare consequences. In to be driven by omitted variables, as any potential omitted developing countries, infants, in particular girls, are more variables would have to interact with both the gender of the likely to die when there is a negative economic shock. child and the direction of the income shock. It is hard to Aggregate macroeconomic shocks involve both income imagine what such an omitted variable would be.16 and substitution effects for individual households. Given positive income gradients in child health, the income effect 16 would generally result in an increase in mortality. But In the working paper version of our paper (Baird, Friedman, & Schady, 2007), we also show that our results are insensitive to the inclu- there is also a substitution effect, as economic shocks sion of controls for a number of possible omitted variables such as rain- decrease the opportunity cost of time and may free up fall; conflict, including civil war; and inflation and other measures of the mothers for time-intensive tasks that have positive effects quality of governance. This also suggests that our estimates of the effect of aggregate income shocks on Infant Mortality Rate are not driven by the on child health—for example, collection of clean water, omitted variables that have received the most attention in the empirical preparation of food, or regular visits to health centers. The literature. The observed asymmetry by gender also makes it very unlikely that our results are driven by recall bias in the DHS. For example, in prin- effect of aggregate economic contractions on child health ciple, one might be concerned that mothers use a salient event like an eco- is therefore hard to sign ex ante. The literature on the Uni- nomic crisis to date an infant death, which could induce a spurious asso- ted States suggests that child health generally improves, ciation between negative income shocks and mortality, It seems unlikely, however, that this sort of recall bias would be present with female deaths and infant mortality declines, during economic contrac- but not male deaths. tions. AGGREGATE INCOME SHOCKS AND INFANT MORTALITY IN THE DEVELOPING WORLD 855 Our results suggest that economic shocks in the develop- reduce the volatility of per capita GDP in developing coun- ing world generally lead to more infant deaths, especially tries or that protect health status during sudden economic of girls, and especially when these shocks are severe. The downturns may have significant benefits for child survival, difference with the findings for the United States, where especially that of girls. infant mortality appears to be procyclical (Chay & Green- stone, 2003; Dehejia & Lleras-Muney, 2004) is striking but REFERENCES perhaps not unexpected. Economic shocks in the develop- Aguiar, Mark, and Gita Gopinath, ‘‘Emerging Market Business Cycles: ing world are often much deeper than those experienced in The Cycle Is the Trend,’’ Journal of Political Economy 115 developed countries, and it is the largest (negative) shocks (2007), 69–102. that have the most serious consequences for infant mortal- Almond, Douglas, Lena Edlund, and Ma ˚ Palme, ‘‘Chernobyl’s Subclinical Legacy: Prenatal Exposure to Radioactive Fallout and School Out- ity. Moreover, developing countries are, by definition, comes in Sweden,’’ Quarterly Journal of Economics 124 (2009), poorer than developed countries, and we show that within 1729–1772. developing countries, the biggest effects of shocks on infant Anderson, Frank, Sarah Morton, Sujata Naik, and Betle Gebrian, ‘‘Maternal Mortality and the Consequences on Infant and Child mortality occur in the poorest countries. Survival in Rural Haiti,’’ Maternal and Child Health Journal 11 Even within developing countries, there is variation in (2007), 395–401. the effects of aggregate economic shocks on infant mortal- Baird, Sarah, Jed Friedman, and Norbert Schady, ‘‘Infant Mortality over the Business Cycle in the Developing World,’’ Policy Research ity. Miller and Urdinola (2010) argue that decreases in the working paper no. 4346 (2007). price of coffee in Colombia, and the attendant reduction in Behrman, Jere, and Anil B. Deloalikar, ‘‘Health and Nutrition,’’ (vol. 1, income in coffee-growing areas, led to reductions in infant pp. 631–711), in Hollis Chenery and T. N. Srinivasan (Eds.), Handbook of Development Economics (Amsterdam: Elsevier, mortality. In their review, Ferreira and Schady (2009) con- 1988). trast the cases of crises in Indonesia in the late 1990s and Bhalotra, Sonia, ‘‘Fatal Fluctuations? Cyclicality in Infant Mortality in Peru in the late 1980s. They argue that the much larger India,’’ Journal of Development Economics 93 (2010), 7–19. Chay, Kenneth Y., and Michael Greenstone, ‘‘The Impact of Air Pollution increase in infant mortality in Peru than in Indonesia may on Infant Mortality: Evidence from Geographic Variation in Pollu- have been a result in part of the protection of health expen- tion Shocks Induced by a Recession,’’ Quarterly Journal of Eco- ditures in Indonesia (but not in Peru). We cannot systemati- nomics 118 (2003), 1121–1167. Cutler, David, Felicia Knal, Rafael Lozano, Oscar Mendez, and Beatriz cally explore these differences with the data at hand. Never- Zurita, ‘‘Financial Crisis, Health Outcomes and Aging: Mexico in theless, our results suggest that the findings from a handful the 1980s and 1990s,’’ Journal of Public Economics 84 (2002), of country-specific studies, including Cutler et al. (2002) on 279–303. Cutler, David, Angus Deaton, and Adriana Lleras-Muney, ‘‘The Determi- Mexico, Paxson and Schady (2005) on Peru, and Bhalotra nants of Mortality,’’ Journal of Economic Perspectives 20 (2006), (2010) on India, hold for a much larger sample of develop- 97–120. ing countries and time periods. We also show that the Deaton, Angus, Global Patterns of Income and Health: Facts, Interpreta- tions, and Policies, NBER working paper no. 12735 (2006). effects of crises on infant mortality appear to be much more Dehejia, Rajeev, and Adriana Lleras-Muney, ‘‘Booms, Busts, and Babies’ severe for girls than boys. Health,’’ Quarterly Journal of Economics 119 (2004), 1091–1130. We conclude by discussing two areas where our data Ferreira, Francisco, and Norbert Schady, ‘‘Aggregate Economic Shocks, Child Schooling, and Child Health,’’ World Bank Research Obser- impose limitations on the possible analysis we can conduct. ver 24 (2009), 147–181. The first of these is the timing of the GDP shocks. Our Gallup, John Luke, and Jeffrey D. Sachs, ‘‘The Economic Burden of Malaria,’’ American Journal of Tropical Medicine and Hygiene 64 results suggest that it is macroeconomic conditions around (2001), 85–96. birth, rather than in the early in utero period or the later half Gwatkin, Davidson R., Shea Rutstein, Kiersten Johnson, Eldaw Suliman, of a child’s first year of life, that matter most for a child’s Adam Jagstaff, and Agbessi Amouzou, Socio-Economic Differ- ences in Health, Nutrition and Population within Developing survival in her first year. However, with annual economic Countries (Washington, DC: World Bank, 2007). data like those we use, it is not possible to tease out the rela- Jamison, Dean T., Martin E. Sandbu, and Jia Wang, ‘‘Why Has Infant tive importance of conditions in narrow windows of expo- Mortality Decreased at Such Different Rates in Different Coun- tries?’’ Disease Control Priorities Project working paper no. 21 sure. Our results on the timing of the shock to aggregate (2004). income should therefore be viewed as suggestive rather Lawn, Joy E., Simon Cousens, and Jelka Zupan, ‘‘Four Million Neonatal than definitive. Second, because we construct birth and Deaths: When? Where? Why?’’ Lancet 365:9462 (2005), 891–900. Loayza, Norman V., Romain Rancie ` re, Luis Serve´ n, and Jaume Ventura, death histories retrospectively, we do not have data on the ‘‘Macroeconomic Volatility and Welfare in Developing Countries: utilization of health services before, during, and after birth An Introduction,’’ World Bank Economic Review 21 (2007), 343–357. for the majority of births (and deaths) we observe. Further, Miller, Grant, and B. Piedad Urdinola, ‘‘Cyclicality, Mortality, and the Value of Time: The Case of Coffee Price Fluctuations and Child the DHS data we use do not include information on other Survival in Colombia,’’ Journal of Political Economy 118 (2010), potential inputs into child health, such as the consumption 113–155. of nutritious foods. We are therefore unable to explore in a Paxson, Christina, and Norbert Schady, ‘‘Child Health and Economic Cri- sis in Peru,’’ World Bank Economic Review 19 (2005), 203–223. comprehensive manner the transmission mechanisms from Preston, Samuel, ‘‘The Changing Relation between Mortality and Level of income shocks to infant mortality. Nevertheless, our results Economic Development,’’ Population Studies 29 (1975), 231–248. clearly indicate that short-term fluctuations in aggregate ——— ‘‘Causes and Consequences of Mortality Decline in Less-Devel- oped Countries during the Twentieth Century’’ (pp. 289–360), in income can have important consequences for the likelihood Richard Easterlin (Ed.), Population and Economic Change in Devel- that a child survives her first year of life. Policies that oping Countries (Chicago: University of Chicago Press, 1980). 856 THE REVIEW OF ECONOMICS AND STATISTICS Pritchett, Lant, and Lawrence H. Summers, ‘‘Wealthier Is Healthier,’’ Strauss, John, and Duncan Thomas, ‘‘Health, Nutrition, and Economic Journal of Human Resources 31 (1996), 841–868. Development,’’ Journal of Economic Literature 36 (1998), 766–817. Rogot, E., P. D. Sorlie, N. J. Johnson, and C. Schmitt, A Mortality Study World Bank, World Development Indicators (Washington, DC: World of 1.3 Million Persons (Bethesda, MD: National Institutes of Bank, 2001). Health, National Heart Lung and Blood Institute, 1992). World Bank. World Development Indicators. http://ddp.worldbank.org/ Ruhm, C. J. ‘‘Are Recessions Good for Your Health?’’ Quarterly Journal ddp/home.do (2007). of Economics 115 (2000), 617–650. World Health Organization, Macroeconomics and Health: Investing in Ruhm, C. J., and W. E. Black, ‘‘Does Drinking Really Decrease in Bad Health for Economic Development, http://www3.who.int/whosis/ Times?’’ Journal of Health Economics 21 (2002), 659–678. cmh (2001). Schultz, Paul T., ‘‘Why Governments Should Invest More to Educate World Health Organization, Neo-Natal and Peri-Natal Mortality: Country, Girls,’’ World Development 30 (2002), 207–225. Regional, and Global Estimates. (Geneva: WHO Press, 2006).