Advertisement

Journal of Economics, Race, and Policy

, Volume 2, Issue 4, pp 202–224 | Cite as

Gender Gaps in Birth Weight Across Latin America: Evidence on the Role of Air Pollution

  • Gabriela Aparicio
  • María Paula GerardinoEmail author
  • Marcos A. Rangel
Original Article
First Online:
  • 191 Downloads

Abstract

Recent estimates indicate that more than 100 million people in Latin America and the Caribbean are exposed to air pollution levels exceeding World Health Organization guidelines. Air pollution persists because of a development process centered around high rates of urbanization and congestion, geographically concentrated industrialization, and biomass burning. This paper focuses on a relatively understudied consequence of this pollution-intensive development process: its gender impact. The analysis provides systematic evidence across the region on the impact of in utero exposure to air pollution on infant health and well-being, a period when the medical literature suggests male fetuses are more delicate than female fetuses. Health at birth is known to have long-term consequences, so this investigation seems warranted and aids the understanding of future gender gaps in socioeconomic development. The empirical analysis combines satellite and survey data from three countries in the region: Bolivia, Colombia, and Peru. Based on sibling comparisons, the analysis finds that a 10% increase in pollution exposure in utero reduces the male–female birth weight gap by approximately 50 g. This weight reduction is equivalent to the impact of smoking five cigarettes a day (versus none) during pregnancy.

Keywords

Air pollution Particulate matter Satellite data Gender gaps Birth weight 

Introduction

Recent estimates indicate that more than 100 million people in Latin America and the Caribbean are exposed to air pollution levels exceeding World Health Organization guidelines (Cifuentes et al. 2005). Though important efforts have been made during the last two decades to improve air quality in the region, air pollution remains high in many large urban centers and is becoming an issue in emerging cities.1 Urban air pollution is primarily the result of burning fossil fuels, and its key sources are transportation, electricity generation, and industry. Nonetheless, pollution levels are also high in certain rural areas of the region, mainly due to deforestation and biomass burning (Provençal et al. 2017). Outdoor air pollution tends to be worse in lower-income communities, affecting precisely those families that have the least resources to protect themselves. Exposure to pollution is also particularly deleterious for those who are most fragile, particularly pregnant women and their fetuses (UNICEF 2016). In utero exposure has been linked to infant mortality, preterm delivery, and low birth weight. Early disadvantages have also been shown to influence long-term trajectories in socioeconomic status. In the present article, we focus on the gender aspects of these disadvantages at birth.

Our study is prompted by recent suggestive epidemiological evidence that male infants are more likely to be adversely affected by air pollution. Van Vliet et al. (2009) document that in over two decades in Canada, the male birth weight advantage shrunk quite significantly. This reduction was approximately half of 1% per year. They maintain that chemical pollutants interfere with hormones and affect males during fetal development more severely than females. Those chemicals with anti-androgenic properties affecting the endocrine system of infant males can attach to aerosols and particulate matter. Our paper extends that research to investigate whether a pattern connecting pollution exposure to gender gaps in birth outcomes can be detected in Latin America. We motivate the analysis presented later in the paper with a simple comparison between the Canadian case and administrative data for Colombia. Interestingly, Fig. 1 shows that the shrinkage in the birth weight gap between males and females is also noticeable for Colombia and is comparable in magnitude to that observed in Canada.
Fig. 1

Difference in mean birthweight by gender: Canada and Colombia. Sources: Van Vliet et al. (2009) for Canada; authors’ calculations based on administrative data for Colombia

For this study, we designed a systematic approach to the data in order to extract comparable information across three countries: Colombia, Bolivia, and Peru. The analysis takes advantage of comparable survey data across countries. Health outcomes are obtained from multiple rounds of the Demographic Health Surveys (DHS) for Bolivia (2008), Colombia (2010 and 2015), and Peru (2012). Our main outcomes of interest are birth weight in grams and an indicator for low birth weight. In the absence of comparable data on ground measures of pollution, the analysis combines these survey data with pollution data from satellites in the form of the aerosol index (AOD) from the Moderate Resolution Imaging Spectroradiometer (MODIS) database of the U.S. National Aeronautics and Space Administration (NASA). AOD is a qualitative measure of aerosols and a good predictor of particulate matter concentration in the air. The data are aggregated at the month level and assigned to each child based on months of exposure in utero conditional on the child’s date of birth and the municipality of residence of the mother during the baby’s gestation.

Aside from the construction of a comparable dataset across Latin American locations, we extend the analysis by focusing on comparisons of pollution exposure across siblings, which allows controlling for confounding factors such as family characteristics and propensity to invest in children that may have blurred previous estimations in the epidemiological literature. Hence, differential effects by gender may be driven by children’s sex-linked physiological differences in the timing of development in utero or hormonal differences rather than by social factors or maternal behavior.2 Our results suggest that there is a negative effect of higher pollution exposure in utero on birth weights, particularly for male babies. The main finding of the study is that a 10% increase in pollution exposure in utero appears to reduce the gender birth weight gap by approximately 50 g. To put this effect into perspective, the impact of smoking five cigarettes a day (versus none) during pregnancy is also 50 g.

This paper is motivated by two main limitations of the current economics literature on the topic. First, few economic studies examine the effects of pollution exposure in utero in the context of developing countries.3 The impact of pollution exposure in utero in developing countries is of independent interest, and it has been hypothesized that the effects of pollution may be larger in developing countries than in developed ones. This may occur, for instance, because pollution levels are generally higher and infant health is often worse in developing countries (Currie et al. 2014). Arceo et al. (2015) is one of the few economics papers to study the effects of pollution exposure in utero in a developing country and compare their findings to those of more advanced economies.4 Interestingly, their estimates of the effects of PM10 on infant mortality for Mexico tend to be similar than estimates for the USA. However, it is not clear whether these findings can be extrapolated to other countries in Latin America.

Second, to the best of our knowledge, there is no paper in the economics literature that has systematically studied the differential effects of in utero pollution exposure by gender. The few available pollution studies with a gender focus in the economics literature have generally looked at outcomes later in life, such as education and labor market outcomes. As is the case of the epidemiological literature, these studies tend to find that adult women may be more adversely affected by pollution than men. Higher vulnerability of women appears to be driven, at least to some extent, by social factors. For example, cognitive damage caused by pollution exposure may affect subsequent human-capital investment decisions for women more than for men (Molina 2016). However, it is not clear whether these findings can be extrapolated to other age cohorts. A more thorough understanding of the effects of air pollution is important to implement policies that effectively address its harmful effects.

The main contribution of the present study to the literature is showing that pollution exposure in utero reduces birth weight notably more for infant males than for females, even when relying on a more rigorous analysis than those commonly used in the epidemiological literature. Our preferred specification controls for family fixed effects, ensuring that results are not affected by family characteristics that remain constant over time, such as socioeconomic status and pollution avoidance preferences. This main result remains qualitatively unchanged across several specifications and robustness checks. Therefore, the study provides supportive evidence that the male–female birth weight gap may be shrinking due to augmented air pollution across the region. We see this study as the first step of an agenda that pursues the analysis of these gender effects in specific contexts and where identifying variations are more suitable to causal inference. We leave that to future research.

Literature Review

The epidemiological literature has long discussed the harmful effects of pollution exposure in utero on human health. However, it is only recently that pollution has become a relevant topic in the economics literature. Two important contributions of this literature to the field are (1) the focus on the role of unobserved heterogeneity and (2) the study of different measures of well-being in addition to more traditional health outcomes.

The Epidemiological Literature

Several epidemiological studies have linked particulate matter and aerosols to adverse health outcomes at all stages of life. Elevated levels of fine particulate matter are associated with increased mortality due to cardiopulmonary diseases, acute respiratory infections, and cancers. Importantly, the harmful effects of air pollutants have been shown to begin in utero, when cells are particularly sensitive to the damage caused by environmental toxins. Ultrafine air pollutants can cross the placental barrier, injuring the developing fetus when the mother is exposed to pollutants. In utero exposure to pollution has been linked to various disorders for the infant, such as mortality, preterm delivery, and low birth weight, among others. As a result, it is now considered just as important for pregnant women to avoid air pollution as it is to avoid smoking (UNICEF 2016).

Epidemiological studies have also highlighted gender disparities in vulnerability to air pollution. Though the literature is far from conclusive, most evidence among adults and older children suggests that females are more severely affected by pollution than males. However, evidence shows that the opposite is true among younger children—early in life, the adverse effects of pollution appear to be stronger among male fetuses and infant boys (Drevenstedt et al. 2008 and Lampl et al. 2009). In terms of pollution, male fetuses may be more at risk due to diverse physiological reasons such as sex-differing lung function growth rates and lower respiratory volumes due to greater airway resistance, among other reasons (Clougherty 2011). Jedrychowski et al. (2009) is one of the few studies to directly assess sex differences in fetal growth reduction in newborns exposed prenatally to fine particulate matter. Focusing on a sample of close to 500 pregnant women, they find larger birth weight and length deficits among males relative to females in response to increased concentration in particulate matter in the air. However, as discussed in the “Introduction,” the epidemiological literature often fails to control for differences in family characteristics that are constant over time, and which may affect the findings.

Economics Literature

Most papers on the effects of exposure to air pollution in utero have focused on very short-term outcomes (such as the effects on infant health) or medium-term effects (such as the well-being of small children). Evidence from developed countries shows that exposure to higher pollution in utero results in (1) higher infant mortality, reduced birth weight, and reduced gestation length; and (2) more hospitalizations for respiratory infections or asthma for small children (see studies cited in Currie et al. 2014).

Particularly relevant for this study is Currie et al. (2009), who use a similar empirical strategy as the one exploited in this paper. They study a large sample of infants born in New Jersey from 1989 to 2006 who were subjected to various levels of pollution in utero. By including family fixed effects, they control for the fixed effects of family background shared by siblings. Their estimates imply that, on average, moving from an area with high carbon monoxide levels to an area with low levels would have a larger effect on infant health than having a pregnant woman reduce her smoking from 10 cigarettes a day to zero. Also relevant for our study is Sanders and Stoecker (2015), which is one of the few studies to consider differential effects by gender. They examine the effects of pollution on sex ratios at birth. Consistent with the hypothesis that male fetuses are more fragile than female fetuses, they find that a reduction in pollution increases the fraction of male fetuses. Lower male ratios are interpreted as an indication of gendered fetal losses.

Studying the effects of pollution in developing countries is important because effects may not be the same as in developed countries. Given that developing countries have lower levels of health, it is plausible that similar levels of pollution may have a much larger health impact. The availability of studies focusing on developing countries is quite limited due to data constraints and the difficulty of finding datasets that allow for linking in utero environmental quality with outcomes.

Some papers have begun to overcome the data difficulties by relying on satellite data and/or by focusing on cohort studies that compare cohorts exposed to some major pollution event to cohorts born just before.5 Jayachandran (2009) uses satellite aerosol measures to track smoke from fires in Indonesia. Her results show a reduction in cohort size for locations exposed to the fires’ smoke during what would be the third trimester of a pregnancy. This suggests that fires resulted in a 20% increase in deaths among fetuses and children less than 3 years of age. Foster et al. (2009) also use satellite measurements to approximate pollution levels throughout Mexico. Using participation in a voluntary pollution reduction program, the authors show that reductions in pollution improve infant mortality from respiratory causes. Also focused on Mexico, Arceo et al. (2015) use thermal inversions, a meteorological phenomenon that traps pollution closer to the ground, and find that PM10 has similar per-unit effects on infant mortality than in the USA. However, CO has larger effects on infant mortality than in the USA. To the best of our knowledge, there are no studies focusing on differential effects by gender of in utero pollution exposure in Latin America.

One of the few available studies that follows individuals over time is Bharadwaj et al. (2017). Using a large dataset of Chilean children, they examine the relationship between air pollution exposure in each month of pregnancy and fourth- and eighth-grade test scores. They find significant effects of exposure to carbon monoxide and ozone in the third and fourth months of pregnancy, timing that is consistent with the results found in Black et al. (2013) and Almond et al. (2009).

Data

Satellite-Based Pollution Data

Though diverse types of air pollutants exist, particulate matter and associated aerosols are among the most widespread and most dangerous. Particulate matter and aerosols are solid or liquid particles suspended in the air. These particles come in many sizes and shapes and can be made up of hundreds of different chemicals. While primary particles are emitted directly from a source (i.e., road dust, smoke from fires, etc.), secondary particles form in complicated reactions in the atmosphere between primary particles and other chemicals (i.e., chemicals emitted from power plants, automobiles, etc.). In addition, particulate matter and aerosols contain a large proportion of black carbon, which has emerged over the last few years as a major contributor to global climate change. Particulate matter is one of the most widely studied types of air pollution, in part due to data availability from ground stations in developed countries and from satellite sources.

Data on air quality was obtained from NASA’s MODIS AOD products, which have been used recently in the economics literature to study the health effects of air pollution and other topics (Foster et al. 2009; Gendron-Carrier et al. 2018; Hansen-Lewis 2018). MODIS daily data are available for the globe since 2000.6 AOD is a well-established proxy for surface air quality, and it has been shown to be highly correlated with ground-based measures of suspended particulate matter (such as PM10 and PM2.5). There is, however, an important conceptual difference between satellite-based AOD and ground-based air quality monitoring data. Satellite-based AOD measures daytime average conditions over a wide area at the specific time the satellite passes overhead; instead, ground-based instruments record conditions at a specific location over several hours. Hence, some divergence between satellite- and ground-based measures is to be expected (for a recent discussion, see Gendron-Carrier et al. 2018).

The main benefit of AOD relative to ground-based air quality measures is that it is available at high temporal frequency (daily in our case) and on a global scale, allowing for consistent measurements across different regions (urban versus rural) and countries. This is particularly important for this paper, as different countries are studied. There are, however, some drawbacks of AOD relative to ground-based measures.7 While ground-based stations monitor particulate matter on the surface, AOD measures are taken higher in the atmosphere and may not accurately reflect the situation on the surface. It is the condition on the surface that matters the most for health effects. In addition, satellite observations only make coarse characterizations of the aerosol type, such as dust, sulfate, smoke, or mixed. The type and chemical composition of particulate matter, and not only the concentration of particles, is likely to affect health outcomes.8 Despite these limitations, AOD is expected to effectively capture large variations in air quality, which is the primary interest of this paper.

MODIS aerosol data products come at different levels of processing. Level 1 (L1) data are generally raw, level 2 (L2) data include minimal processing, and level 3 (L3) data include higher levels of processing by NASA (i.e., data are provided at equal grids). Generally, higher processing levels allow for easier manipulation at the cost of less flexibility. While recent papers have used L2 data, we have opted to use L3 data, as the variability of interest is at the municipality level and very fine geographic measurements are not needed (municipalities are the smallest geographic units that can be matched with health data in demographic health surveys). The specific AOD dataset used in this paper is the MOD09CMA product, which has a spatial resolution of 0.05° (approximately 5 km per pixel) and includes only AOD measures over land (Terra satellite).9 The data have been processed by NASA to account for clouds and high solar zenith angles, and are generally used for climate modeling.10

MODIS products report AOD on a scale from 0 to 5000, which is generally normalized to take values between 0 and 5.11 Rescaled AOD values can be interpreted as being in a log scale from 0 to 5. AOD below 0.1 is considered clear, while the maximum possible of 5 means that sunlight cannot pass through the air. Since AOD is already on a log scale, additional log transformation of the data is unnecessary (Hansen-Lewis 2018).

Pollution Patterns in Latin America

We provide a simple validation of the use of AOD data as a pollution proxy for three selected cities. Figure 2 compares satellite-based AOD with ground-based pollution readings from the World Health Organization’s (WHO) Global Urban Ambient Air Pollution Database.12 The WHO database collects information on average annual PM10 and PM2.5 concentrations (mg/m3) for cities around the world where ground stations are available.13 Figure 2 suggests that broad characterizations of pollution levels across cities, based on AOD and PM10, are qualitatively comparable. Considering either AOD or PM10, Lima is by far the most polluted city (among the three being studied), followed by Bogota and then La Paz. Importantly, all three cities have PM10 levels above the maximum annual average exposure level of 20 mg/m3 recommended by the WHO and AOD levels exceeding the 0.1 threshold of clean air suggested in Hansen-Lewis (2018). Despite the conceptual difference between AOD and ground-based measures, the similar patterns observed are not surprising. As discussed earlier, research in atmospheric science has validated MODIS AOD as a measure of ground-level fine particulate matter globally as well as in several countries (Gendron-Carrier et al. 2018).14
Fig. 2

Comparison of aerosol index (AOD) and PM10: Bogota, La Paz, and Lima. Source: Prepared by the authors. Note: The calculation of city-level aerosol index (AOD) measures relies on the official geocodes used in each country. La Paz is identified based on municipality code 20101, Bogota is identified based on municipality code 11001, and Lima is identified based on district code 150101

Figure 3 describes broad trends in mean AOD over time for selected countries and cities in Latin America between 2000 and 2015. AOD has remained quite flat over time when looking at country-level aggregates. Studies of pollution in Latin America confirm that for the entire region, as well as for most individual cities, pollution has remained constant or is decreasing slightly (Provençal et al. 2017). However, the current situation in Latin America is still worrisome because pollution has remained at relatively high levels. On average, Bolivia, Colombia, and Peru have AOD levels that are considered polluted, as they are above the 0.1 threshold. Moreover, Fig. 3 presents national averages and, consequently, may be concealing much variation in pollution that occurs within different municipalities.
Fig. 3

Pollution trends: selected countries and cities. Source: Prepared by the authors. Note: AOD, aerosol index

A thorough discussion of pollution trends over time is particularly important in the context of our paper, as these data are employed to relate the variation of pollution within a given municipality/district over time with birth outcomes. Figure 4 shows a histogram of over 500,000 municipality–month pollution observations available for Bolivia, Colombia, and Peru over 2000–2015.15 It can be seen that municipality–month pollution levels take a number of different values, ranging from 0 to 0.4, with most of the distribution concentrated above the 0.1 threshold for clean air. This wide range in values suggests that there is important variability in pollution across locations. Figure 5 shows the pollution growth rate across time for different groups of municipalities. Municipalities are classified as high or low pollution growth (relative to the median municipality). Both groups have very different patterns in pollution over time, which suggest that pollution growth is heterogeneous within municipalities. Furthermore, even aggregate pollution levels need to be measured on a population-weighted basis. Since more people live in urban polluted centers, it is expected that the aggregate weighted measure will have a different trend. Figure 6 shows that in a quarter of municipalities, pollution grew by over 1% between 2001 and 2014.
Fig. 4

Histogram of municipality–month pollution levels. Source: Prepared by the authors. Note: AOD, aerosol index

Fig. 5

Pollution trends by country. Source: Prepared by the authors

Fig. 6

Distribution of the growth rate of municipalities. Source: Prepared by the authors

Further evidence of the variation of pollution over time within each location is shown in Fig. 7, which presents box plots of the distribution of monthly pollution levels for Bolivia, Colombia, and Peru. The box plots show that there is important variability over time in pollution within a country, even though long-term pollution trends are quite stable.16 When looking at specific cities, the variability in pollution over time is somewhat smaller but remains significant for some cities such as Lima. There are many reasons for pollution levels to vary within a year in a given location. For example, there may be less congestion during months when schools are in recess and, consequently, lower pollution levels. Similarly, industries that emit pollution may be more active in certain months than others. Climatic factors may also play a role. Winds that move pollution into or out of a given location may vary in strength across different months. Similarly, the likelihood of inversion episodes that trap pollution in a given location also varies over time.
Fig. 7

Box plots of pollution variation within selected countries and cities. Source: Prepared by the authors. Note: AOD, aerosol index

Demographic Health Survey Data

Infant health outcomes were obtained from several rounds of demographic health surveys, which provide comparable data for several countries. Specifically, this paper focuses on DHS data for Bolivia (2008), Colombia (2005 and 2010), and Peru (2012). We consider boys and girls in the DHS birth roster (those born up to 5 years before the time of the survey). The main outcomes of interest are birth weight measures. Other relevant information is also collected in the survey, such as demographics and other household characteristics.

Descriptive statistics are shown in Table 1. Panel A focuses on our main outcome of interest. It shows the distribution of the birth weight of children by country and gender. On average, boys weigh around 100 g more than girls in all the countries. On average, children are heavier in Bolivia and lighter in Peru. The mean birth weight for girls varies between 3271 and 3155 g, and for boys between 3393 and 3265 g. Panel B displays additional summary statistics of child characteristics by country. In the sample, child characteristics across countries are very similar. On average, children were about 2 years of age when the survey was conducted, and 51% were male. Less than 10% of children were born with low birth weight in the three countries.
Table 1

Summary statistics from demographic health surveys

Panel A

Birth weight

Percentiles

Bolivia

Colombia

Peru

Girls

Boys

Girls

Boys

Girls

Boys

5

2400

2500

2250

2300

2200

2300

25

3000

3000

2800

2996

2800

2950

50

3300

3400

3200

3300

3200

3280

75

3600

3800

3500

3600

3500

3620

95

4100

4500

4000

4200

4000

4120

Mean

3271

3393

3175

3283

3155

3265

Std. dev

557

599

592

613

559

581

Observation

3101

3357

9477

9923

12,485

12,815

Panel B

Variables

Observations

Mean

Standard deviation

Minimum

Maximum

Bolivia

  Low birth weight

6458

0.05

0.23

0

1

  Pollution

6458

0.15

0.06

0.05

0.37

  Current age of child

6277

1.94

1.41

0

4.00

  Male child

6458

0.52

0.50

0

1

  Child alive

6458

0.97

0.17

0

1

  Number of children under age 5

6458

1.61

0.79

0

6.00

  Total children ever born

6458

3.11

2.21

1

14.00

  Nonmigrant

6457

0.87

0.33

0

1

Colombia

  Low birth weight

19,370

0.08

0.27

0

1

  Pollution

19,004

0.19

0.04

0.07

0.41

  Current age of child

19,167

1.62

1.34

0

4.00

  Male child

19,370

0.51

0.50

0

1

  Child alive

19,370

0.99

0.10

0

1

  Number of children under age 5

19,370

1.53

0.77

0

7.00

  Total children ever born

19,370

2.32

1.54

1

14.00

  Nonmigrant

19,370

0.84

0.36

0

1

Peru

  Low birth weight

25,300

0.08

0.27

0

1

  Pollution

25,280

0.16

0.05

0.03

0.79

  Current age of child

24,891

2.02

1.40

0

4.00

  Male child

25,300

0.51

0.50

0

1

  Child alive

25,300

0.98

0.13

0

1

  Number of children under age 5

25,300

1.45

0.71

0

7.00

  Total children ever born

25,300

2.80

1.90

1

15.00

  Nonmigrant

25,292

0.89

0.31

0

1

Source: Prepared by the authors

Matched Demographic Health Survey Pollution Data

Demographic health surveys include geographic information that can be used to merge the health and demographic data with the AOD pollution proxy discussed earlier. We rely on the official municipality codes for Bolivia and Colombia and district codes for Peru to merge the data. Both municipalities and districts are equivalent third-level administrative units. First, pollution data were aggregated at the monthly level by choosing the median value occurring in a given month for each 5-km pixel.17 Then, municipality–month pollution levels were obtained by calculating the municipality average, for each given month, of all the monthly level pixels that fall within the geographic boundaries of a given municipality/district. Nine municipality–month pollution measures were assigned to each child in the birth records of the DHS conditional on their date of birth and the place of residence of the mother during pregnancy. Given that the duration of the pregnancy is endogenous, in all cases, it is assumed that the gestation period lasted the full 9-month term. Each child’s average in utero pollution exposure is therefore obtained by averaging the municipality–month pollution levels for the 9 months corresponding to the approximated gestation period.

Panel B of Table 1 also describes exposure to pollution in utero specifically for children in the DHS. The average pollution exposition in utero is 0.15 in Bolivia, 0.16 in Peru, and 0.19 in Colombia. Also, there is a lot of variation within countries. Even though on average Colombia seems to be where children are more exposed to pollution, Peru has the highest level of municipality-pregnancy pollution exposure (0.79).

This spatiotemporal variation in pollution exposure in pregnancies reported in the DHS can be explored in further detail. The municipal averages of the in utero pollution exposure of children identified in the DHS are presented in Figs. 8, 10, and 12 for Bolivia, Colombia, and Peru, respectively.18 The figures show that there is geographical variation in pollution and that pollution is distributed as expected. For example, in Colombia, pollution is concentrated in the more urban west, rather than in the Amazon in the east. In addition, the figures also show that pollution levels in the countries of interest are relatively high. The colors in the figures suggest that most municipalities exhibit pollution levels above the threshold defining clean air (pollution levels under 0.1 are considered clean air). The variation in pollution within municipalities is presented in Figs. 9, 11, and 13 for Bolivia, Colombia, and Peru, respectively. Specifically, these maps show the interquartile range of in utero pollution exposure of children in the DHS survey by municipality. There is large across-municipality and  within-municipality variation in in utero pollution exposure (Figs. 8, 9, 10, 11, 12, and 13).
Fig. 8

Municipal average of in utero pollution exposure of children in demographic health survey, Bolivia. Source: Prepared by the authors

Fig. 9

Municipal interquantile range of in utero pollution exposure of children in demographic health survey, Bolivia. Source: Prepared by the authors

Fig. 10

Municipal average of in utero pollution exposure of children in demographic health survey, Colombia. Source: Prepared by the authors

Fig. 11

Municipal interquantile range of in utero pollution exposure of children in demographic health survey, Colombia. Source: Prepared by the authors

Fig. 12

Municipal average of in utero pollution exposure of children in demographic health survey, Peru. Source: Prepared by the authors

Fig. 13

Municipal interquantile range of in utero pollution exposure of children in demographic health survey, Peru. Source: Prepared by the authors

Methodology

Baseline Estimation

The baseline empirical strategy exploits the time–municipality variation in pollution. We empirically test the relationship between the average level of pollution in a municipality during the 9 months of pregnancy and different health outcomes. To account for a potential mismeasurement in pollution exposure, we focus on children whose families have lived in the same location since before the child was conceived.19

For this purpose, the following equation is specified:

$$ {y}_{i,m,t}={\beta}_1{Average\ Pollution}_{i,m,t}+{\beta}_2{Male}_i+{\partial}_{\zeta }+{\mu}_{\mathrm{b}}+{\tau}_m+{\varepsilon}_{i,m,t}, $$
(1)
where y is the health outcome variable for child i, in municipality m and time of birth t, which we also represent by the year of birth (ζ) and month of birth (b). The term Average Pollution denotes the average pollution to which a child is exposed while being in utero during the 9 months prior to delivery. Malei is equal to 1 if the child is a boy and 0 otherwise. μb corresponds to month of birth fixed effect and ∂ζ to year of birth fixed effect that accounts for seasonality differences and yearly trends common to all counties. τm represents municipality fixed effects that control for all time-invariant characteristics varying at the municipality level. εi, m. t is the error term. Because, we control for municipality fixed effects, the coefficient that captures the effect of pollution (β1) is identified using changes within municipalities over time.

One potential concern with the estimation proposed in Eq. 1 is that family characteristics that influence health at birth may be related to specific pollution exposure within a municipality. Therefore, our preferred specification controls for household fixed effects instead of municipality fixed effects (Eq. 2), where τf identify mothers in the sample:

$$ {y}_{i,f,t}={\beta}_1{Average\ Pollution}_{i,f,t}+{\beta}_2{Male}_{i,f}+{\partial}_{\zeta }+{\mu}_{\mathrm{b}}+{\tau}_f+{\varepsilon}_{i,m,t}. $$
(2)

Controlling for household fixed effects is crucial given that children from more disadvantaged households tend to have lower health and educational outcomes. In addition, household fixed effects may also help to control for some of the determinants for the decision to move to different locations (locational sorting). Given that household characteristics may be correlated with the decision to migrate, controlling for household characteristics may help to address some of the biases arising from households that choose to relocate. The effects of interest will be identified by the variation in pollution across time, which results in siblings being exposed to different pollution levels. This differences out any family-specific or municipality characteristics that affect children across families. This fixed-effect model looks at the differences on health within families residing in municipalities with different levels of pollution over time and that have more than one child.

Heterogeneous Effects by Gender

To examine the main question posed in this study, this section includes an interaction term between the measure of pollution exposure and the gender of the child (Eq. 3):

$$ {\displaystyle \begin{array}{rcl}{y}_{i,f,t}& =& {\beta}_1{Average\ Pollution}_{i,f,t}+{\beta}_2{Male}_{i,f}+\\ {}& & {\beta}_3{Average\ Pollution}_{i,f,t}\times {Male}_{i,f}+{\partial}_{\zeta }+{\mu}_{\mathrm{b}}+{\tau}_f+{\varepsilon}_{i,m,t}.\end{array}} $$
(3)

This specification serves to examine whether changes in pollution have an impact on the birth outcome, which is a function of the child’s gender. As a result, and according to Eq. (3), the main coefficient of interest is the interaction term between the pollution variable and the male dummy (β3). This implies that the expected change in the gender gap in the birth outcome, due to an increase in pollution of δ, will be given by (0.01 ∗ β3 ∗ δ), as pollution is measured in logs.20 As before, this household fixed-effect model looks at the differences in health outcomes within families that reside in municipalities with different levels of pollution across time and that, at a minimum, have two children.

The analysis conducted controls for the potential bias arising from migration. To this end, the main results presented in this paper focus on a sample of children whose mother has been living in the same municipality since before the child was conceived (we refer to this sample as the nonmigrant sample). Though this is a selected sample, it reduces noise by ensuring that each child is assigned the level of pollution that corresponds to the municipality where the mother resided while expecting the child. We present results for the full sample, including both migrant families (those that have moved to a different municipality since the child was born) and nonmigrant families as robustness checks. Results are qualitatively similar regardless of the sample studied. Hence, we discuss only the results for the nonmigrant sample.

Main Results

We study the effects of pollution on two main outcomes: (1) birth weight (measured in grams) and (2) an indicator of whether the baby has low birth weight. The birth weight indicator is a common and very important outcome used in this literature (Currie et al. 2009). Low birth weight is one of the crucial factors affecting child morbidity and mortality worldwide; approximately one third of neonatal deaths are attributable to it. The WHO defines low birth weight as the weight of live born infants of less than 2500 g, regardless of gestational age or any other etiology. Unfortunately, demographic health surveys do not record other indicators of health at birth, such as prematurity or APGAR scores.

Baseline estimates are presented in Table 2. Four models are presented, with each subsequent model controlling for additional confounding factors. In all cases, we control for month of birth fixed effects because seasonality is known to have an influence on both fertility and pollution and even potentially direct effects on birth outcomes. Also, we control for year of birth fixed effects to account for any overall trends in pollution and health at birth across the continent. Additionally, column (1) controls for survey fixed effects to account for differences in the field work across surveys. Column (2) controls for municipality fixed effects to account for differences across municipalities in terms of invariant characteristics. Column (3) includes department–trimester–year of birth fixed effects to control not only for differences across departments (main geographical and administrative division within each country), but also for the possibility of different cyclical trends over time in the departments. Finally, column (4) replaces municipality fixed effects with household fixed effects to account for differences in household characteristics, such as each family’s preferences for pollution exposure and investment in children, as well as differences in socioeconomic status that affect food choices and availability. Additionally, we include month–year fixed effects to account for any time-specific overall trend. In all models, we cluster standard errors at the municipality level.
Table 2

Effect of pollution exposure in utero on birth weight (nonmigrants)

 

(1)

(2)

(3)

(4)

Weight (g)

Weight (g)

Weight (g)

Weight (g)

Variables

  Pollution

596.2*** (100.0)

− 136.3 (126.9)

− 126.4 (166.2)

10.61 (289.2)

  Male

113.6*** (4.982)

115.3*** (4.997)

117.8*** (5.395)

117.7*** (10.59)

  Constant

3459*** (31.60)

3508*** (22.65)

4180*** (331.9)

3356*** (378.2)

Observations

46,851

46,851

46,851

13,319

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.018

0.012

0.063

0.049

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01

Estimates of the association between pollution exposure in utero and birth weight (measured in grams) are shown in Table 2. Column (1) suggests that a 10% increase in pollution exposure in utero is associated with a lower birth weight of more than 50 g.21 However, when controlling for municipality fixed effects (column 2), department–trimester–year fixed effects (column 3), or family fixed effects (column 4), the effects of pollution are no longer statistically significant at conventional levels. Table 3 follows the format of Table 2, but the dependent variable is an indicator for low birth weight rather than weight in grams. Results are qualitatively similar to those obtained earlier.
Table 3

Effect of pollution exposure in utero on an indicator for low birth weight (nonmigrants)

 

(1)

(2)

(3)

(4)

Low birth weight

Low birth weight

Low birth weight

Low birth weight

Variables

  Pollution

0.0949*** (0.0307)

0.0504 (0.0491)

0.0624 (0.0683)

− 0.0973 (0.116)

  Male

0.0185*** (0.00249)

0.0191*** (0.00242)

0.0194*** (0.00251)

0.0208*** (0.00614)

  Constant

3459*** (31.60)

3508*** (22.65)

4180*** (331.9)

3356*** (378.2)

Observations

46,851

46,851

46,851

13,319

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.004

0.002

0.050

0.025

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01

There are different reasons why our aggregate results may be inconclusive. One possibility is that pollution has a limited effect on birth weights in developing countries, on average. Another possibility is that the lack of robustness of the results is driven by limited power due to the small number of observations available in our data. While some of the research on this topic in developed countries relied on administrative data where many children are studied, this paper relies on survey data with a more limited number of observations. Survey data were used to ensure comparability across countries and due to the limitations of administrative data in the region. For instance, in most cases, it is not possible to identify siblings based on administrative data.

It is important to consider, however, whether the results in Tables 2 and 3 may be hiding the heterogeneous effects we are interested in. We need to take a closer look at the possibility of differential sensitivities to pollution by gender, as suggested by the epidemiology literature. Relying exclusively on the results presented thus far, which fail to consider the gender perspective, and not exploring the effects of pollution any further may lead to incomplete conclusions and potentially flawed policy recommendations. Pollution exposure for pregnant women may still be a very important concern if it affects some types of babies, even if evidence for the aggregate is inconclusive. Hence, in the remaining portion of this paper, we study the effects of pollution on boys and girls, which informs us about the impact of pollution on birth weight gender gaps.

Table 4 presents estimates of the heterogeneous effects by gender of pollution exposure in utero on birth weight. The main difference between Tables 2 and 4 is that the latter includes an interaction term between male births and pollution (as shown in Eq. 3). As discussed in the literature review, male fetuses are more delicate than female fetuses due to differences in physiological development. Hence, it is reasonable to expect pollution to have differential effects along that dimension.22 In all models, the interaction term between males and pollution has a negative sign (statistically significant at conventional levels), suggesting that pollution is more harmful for males. Most importantly, even when controlling for family fixed effects, higher pollution is associated with lower birth weights for males. A 10% increase in pollution is associated with a reduction in the birth weight gap between males and females of between 20 and 50 g, depending on the specification.23 Table 5 is similar to Table 4, but the dependent variable of interest is low birth weight. Once again, results are qualitatively similar regardless of which of the two outcomes is used. Our preferred specification, which controls for mother fixed effects, suggests that a 10% increase in pollution increases the probability of low birth weight by around 2 percentage points more for boys than for girls.
Table 4

Effect of pollution exposure in utero on birth weight, gender interaction (nonmigrants)

 

(1)

(2)

(3)

(4)

Weight (g)

Weight (g)

Weight (g)

Weight (g)

Variables

  Pollution

708.1*** (107.7)

− 8.945 (134.5)

− 5.483 (170.8)

266.5 (287.6)

  Male

151.4*** (18.73)

158.5*** (19.05)

158.4*** (19.87)

200.2*** (37.23)

  Pollution * male

− 220.5** (103.4)

− 252.1** (104.8)

− 237.2** (108.2)

− 487.8** (213.8)

  Constant

3435*** (33.78)

3481*** (25.11)

4154*** (331.8)

3305*** (375.8)

Observations

46,851

46,851

46,851

13,319

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth fixed effect

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.018

0.012

0.063

0.049

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01; **p < 0.05

Table 5

Effect of pollution exposure in utero on an indicator for low birth weight, gender interaction (nonmigrants)

 

(1)

(2)

(3)

(4)

Low birth weight

Low birth weight

Low birth weight

Low birth weight

Variables

  Pollution

− 0.131*** (0.0388)

0.00573 (0.0536)

0.0170 (0.0712)

− 0.201* (0.119)

  Male

0.0307*** (0.00791)

0.0342*** (0.00790)

0.0346*** (0.00833)

0.0542*** (0.0199)

  Pollution * male

0.0714 (0.0445)

0.0884** (0.0443)

0.0891* (0.0461)

0.198* (0.106)

  Constant

0.00297 (0.0117)

− 0.0143 (0.0107)

− 0.126 (0.122)

0.0747 (0.210)

Observations

46,851

46,851

46,851

13,319

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.004

0.002

0.050

0.026

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01; **p < 0.05; *p < 0.1

In order to put the magnitude of our results into perspective, we compare the effects of in utero pollution exposure to the effects of smoking during pregnancy. Many epidemiology studies find very large reductions in the birth weight of babies of mothers who smoke. After controlling for mother fixed effects, Currie et al. (2009) find that the mother being a smoker reduces the child’s birth weight by approximately 39 g, and each additional cigarette reduces it by a further 2.2 g, for a total reduction of 50 g in infants of women who smoke five cigarettes a day. Hence, a 10% increase in pollution, as measured by the AOD index, has as similar impact on the gender birth weight gap as smoking five cigarettes a day has on the birth weight of infant children. The magnitude of the effects presented is substantive, suggesting that the differential harm of pollution exposure in utero is very sizable. Based on these results, the recommendation by UNICEF (2016) that it is just as important for pregnant women to avoid pollution exposure as it is to avoid smoking seems warranted—particularly for mothers expecting boys.

As discussed in the “Introduction,” there is still a limited understanding of the effects of pollution on the gender birth weight gap both in the economics and epidemiological literature. However, the scarce epidemiological literature available suggests that our results indicating that male fetuses are hurt most by pollution seem reasonable. There is evidence that the birth weight gap between boys and girls is generated by androgen action, and some pollutants have anti-androgenic properties. Boys with androgen-insensitive syndrome have been found to have a comparable weight to girls (De Zegher et al. 1998), and anti-androgenic pollutants may be having a similar effect. Further, evidence shows that some critical time windows of development may be slightly different in boys and girls, with boys developing earlier (De Zegher et al. 1999), and these differences in timing may also play a role in our findings.

Table 6 studies whether higher pollution is associated with lower male births as suggested by some studies in developed countries, being indicative of differential mortality. Additionally, miscarriages of a particular gender due to pollution might introduce selection bias in our analysis. However, we find no statistically significant effects on this variable. Furthermore, when plotting the gender dummy on pollution (using the raw data), we do not see a correlation between changes in the sex ratio and pollution (Fig. 14). We find no clear evidence of sex ratios at birth being correlated with pollution.
Table 6

Effect of pollution exposure in utero on gender (nonmigrants)

 

(1)

(2)

(3)

(4)

Male

Male

Male

Male

Variables

  Pollution

− 0.0137 (0.0483)

0.0667 (0.114)

0.160 (0.141)

0.432 (0.282)

  Constant

1.018*** (0.0177)

0.955*** (0.0209)

0.27 (0.405)

1.065** (0.447)

Observations

46,851

46,851

46,851

13,319

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.000

0.000

0.046

0.025

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01; **p < 0.05

Fig. 14

Sex ratio and pollution. Source: Prepared by the authors

Robustness Checks

This section presents several robustness checks for our main results on the effects of pollution exposure in utero on the gender birth weight gap. The different robustness checks support the findings in Tables 4 and 5 that pollution appears to be shrinking the gender gap in birth weight, in disfavor of boys.

Placebo Test

We conduct placebo tests to show that our main findings regarding the gender birth weight gap are unlikely to be driven by factors other than pollution. For this purpose, we replace the actual 9-month average in utero pollution exposure with a random average exposure. This is done by reorganizing the actual municipality–month pollution levels in random order. This ensures that the random pollution values are drawn from the same distribution as the real pollution values. Tables 7 and 8 present the results of a Monte Carlo simulation of 1000 replications of random order pollution exposures on weight and low birth weight, respectively. Columns (1) to (3) correspond to simulations controlling for month of birth fixed effects and municipality–year fixed effects. Columns (4) to (6) show results when including mother fixed effects. Columns (1) and (4) show that the mean of the coefficients after 1000 simulations is close to zero for both outcomes and very small in magnitude relative to the effect estimated using the real exposure measure. Columns (3) and (6) show the rejection rate calculated as the share of repetitions where the null hypothesis of the effect is rejected at the 5% level. This rejection rate is very low independent of the control variables used and the outcome variables. Finally, we present the results of the placebo test graphically (see Fig. 15) showing the distribution of the placebo coefficient estimates for the interaction between pollution exposures and being a male after 1000 replications. The vertical red line at the left of the figure displays the estimated value of the interaction term based on the actual exposure (equal to − 500 prior to rescaling). There is evidence of a large difference in magnitude of the simulation result and the actual coefficient.
Table 7

Monte Carlo simulations of placebo tests (low birth weight)

Weight

(1)

(2)

(3)

(4)

(5)

(6)

B_Pollution * male

_SE

Rejection rate

B_Pollution * male

_SE

Rejection rate

Mean

0.319

29.814

0.044

− 1.061

59.606

0.063

Standard deviation

29.377

0.576

0.205

60.431

1.060

0.243

Minimum

− 104.520

28.104

0

− 214.025

55.944

0

Maximum

89.433

31.605

1

186.757

63.484

1

Control variables

  First-born

Yes

Yes

Yes

Yes

Yes

Yes

  Month of birth fixed effect

Yes

Yes

Yes

Yes

Yes

Yes

  Municipality * year-birth fixed effect

Yes

Yes

Yes

No

No

No

  Mother fixed effect

No

No

No

Yes

Yes

Yes

Repetitions

1000

     

Source: Prepared by the authors

Table 8

Placebo random pollution exposure (low birth weight)

Low_birth_weight

(1)

(2)

(3)

(4)

(5)

(6)

B_Pollution * male

_SE

Rejection rate

B_Pollution * male

_SE

Rejection rate

Mean

0.001

0.014

0.041

0.001

0.031

0.055

Standard deviation

0.013

0.000

0.198

0.032

0.001

0.228

Minimum

− 0.038

0.013

0

− 0.094

0.028

0

Maximum

0.039

0.015

1

0.095

0.034

1

Control variables

  

  First-born

Yes

Yes

Yes

Yes

Yes

Yes

  Month of birth fixed effect

Yes

Yes

Yes

Yes

Yes

Yes

  Municipality * year-birth fixed effect

Yes

Yes

Yes

No

No

No

  Mother fixed effect

No

No

No

Yes

Yes

Yes

Repetitions

1000

     

Source: Prepared by the authors

Fig. 15

Monte Carlo simulation of placebo test: placebo test versus estimated effect with actual data. Source: Prepared by the authors

Other Robustness Checks

Additional robustness checks are presented in Tables 9 and 10 for birth weight in grams and an indicator for low birth weight, respectively. These tables are similar to Tables 4 and 5, but they also control for whether a child was the first-born. This is an important control, particularly for the specifications comparing the effects of pollution for siblings (column 4), as there is substantive evidence that the birth weight of the second-born child is significantly higher than that of first-born child. For instance, Bacci et al. (2014) find that the birth weight of first-born children is 89 g lower than that of second-born children on average, but the magnitude of the effect also varies by gender. Additionally, mothers may have different attitudes and behaviors conditional on the order of birth of their children. To control for this potential confounding factor, we include a dummy if a child was the first child born to a given mother. As suggested by the literature, we find that the birth weight of the first-born child is, on average, between 30 and 85 g lower than that of other children. However, including this additional control does not change our main findings. The estimates remain statistically significant and magnitudes are like those in the main tables.
Table 9

Controlling for first-born child (weight)

 

(1)

(2)

(3)

(4)

Weight

Weight

Weight

Weight

Variables

  Pollution

149.6*** (18.63)

156.2*** (18.96)

156.1*** (19.76)

198.2*** (37.30)

  Male

717.2*** (108.5)

− 12.89 (133.7)

− 14.58 (170.5)

257.8 (286.8)

  Pollution * male

− 212.4** (103.0)

− 241.1** (104.3)

− 226.7** (107.6)

− 478.0** (214.4)

  First child

− 79.29*** (5.545)

− 85.98*** (5.369)

− 85.53*** (5.505)

− 28.87* (14.94)

  Constant

3421*** (33.74)

3470*** (25.00)

4157*** (309.4)

3354*** (371.6)

Observations

46,851

46,851

46,851

13,319

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.023

0.018

0.069

0.050

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01; **p < 0.05; *p < 0.1

Table 10

Controlling for first-born child (low birth weight)

 

(1)

(2)

(3)

(4)

Low birth weight

Low birth weight

Low birth weight

Low birth weight

Variables

  Pollution

− 0.132*** (0.0389)

0.00617 (0.0536)

0.0179 (0.0713)

− 0.199* (0.119)

  Male

− 0.0305*** (0.00791)

− 0.0340*** (0.00790)

− 0.0344*** (0.00832)

− 0.0537*** (0.0200)

  Pollution * male

0.0705 (0.0445)

0.0872** (0.0443)

0.0880* (0.0461)

0.195* (0.107)

  First child

0.00871*** (0.00278)

0.00938*** (0.00274)

0.00918*** (0.00285)

0.00787 (0.00804)

  Constant

0.00450 (0.0116)

− 0.0131 (0.0107)

− 0.126 (0.120)

0.0614 (0.211)

Observations

46,851

46,851

46,851

13,319

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.004

0.003

0.051

0.026

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01; **p < 0.05; *p < 0.1

Our last robustness checks are presented in Tables 11 and 12. The main difference between these tables and Tables 4 and 5 is that we now focus on the full sample of births available in DHS data, rather than on only those in households that did not migrate. Restricting the sample to only households that do not migrate is important to ensure that pollution is adequately assigned to each birth. However, this analysis has the disadvantage that relies on a selected sample. Households that choose to migrate may do so in part due to pollution. As before, results suggest that pollution is associated with a decrease in the gender birth weight gap. However, coefficient estimates are not always statistically significant at conventional levels. Lack of statistical significance could be due in part to the noise introduced by imprecisely assigning the level of pollution to some children, or due to some endogenous household response to pollution that is not captured by our preferred estimates.
Table 11

Effect of pollution exposure in utero on an indicator for birth weight (full sample including migrants)

 

(1)

(2)

(3)

(4)

Weight (g)

Weight (g)

Weight (g)

Weight (g)

Variables

  Pollution

616.2*** (101.7)

− 111.2 (131.8)

− 103.1 (163.6)

167.9 (267.7)

  Male

135.3*** (19.04)

139.1*** (19.14)

137.3*** (20.14)

194.4*** (34.80)

  Pollution * male

− 137.5 (104.9)

− 158.5 (104.9)

− 140.7 (109.8)

− 465.1** (199.4)

  Constant

3453*** (32.03)

3499*** (25.57)

4227*** (346.4)

3495*** (332.5)

Observations

54,074

54,074

54,074

54,074

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.016

0.011

0.054

0.040

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01; **p < 0.05

Table 12

Effect of pollution exposure in utero on an indicator for low birth weight (full sample including migrants)

 

(1)

(2)

(3)

(4)

 

Low birth weight

Low birth weight

Low birth weight

Low birth weight

Variables

Pollution

− 0.115*** (0.0379)

0.0347 (0.0559)

0.0763 (0.0734)

− 0.132 (0.114)

Male

− 0.0254*** (0.00814)

− 0.0277*** (0.00821)

− 0.0260*** (0.00851)

− 0.0511*** (0.0179)

Pollution * male

0.0457 (0.0452)

0.0577 (0.0455)

0.0485 (0.0464)

0.186* (0.0960)

Constant

− 0.000290 (0.0115)

− 0.0205* (0.0111)

0.0748 (0.101)

0.0611 (0.260)

Observations

54,074

54,074

54,074

54,074

Control variables

  Survey fixed effect

Yes

   

  Month of birth fixed effect

Yes

Yes

Yes

Yes

  Year of birth fixed effect

Yes

Yes

Yes

Yes

  Municipality fixed effect

 

Yes

Yes

 

  Mother fixed effect

   

Yes

  Month-birth * year-birth

   

Yes

  Trimester * year-birth * department fixed effect

  

Yes

 

R2

0.003

0.002

0.043

0.022

Source: Prepared by the authors. Note: Clustered standard errors in parentheses (municipality level)

***p < 0.01; *p < 0.1

Conclusion

Recent estimates indicate that more than 100 million people in Latin America and the Caribbean are exposed to air pollution levels exceeding World Health Organization guidelines. However, there is limited rigorous evidence for the region about the impact of air pollution exposure while in utero on infant health and well-being. Moreover, though epidemiological studies have found that male fetuses are more delicate than female fetuses, few papers in the economics literature have studied the differential effects of air pollution on birth weight by gender.

This paper shows that pollution operates to reduce the male–female birth weight gap. Our results suggest that there may be an association between higher pollution exposure in utero and lower birth weights for male babies, who seem to be more delicate than females in utero. The main finding of the paper is that a 10% increase in pollution exposure in utero appears to reduce the gender birth weight gap by approximately 50 g, when pollution is measured by the AOD Index. Findings underscore the importance of looking at the heterogeneous effects of pollution exposure by gender: ignoring the gender perspective may lead to wrong policy implications. Policy recommendations should focus on the pollution impacts of vulnerable groups, and not only the average impact.

Though this paper and other recent papers have contributed to understanding the effects of in utero pollution exposure on well-being, the available research on the topic remains limited for developing countries, particularly those in Latin America. Several opportunities for further research are available. Future work could expand the evidence presented here by studying additional countries. An important advantage of relying on satellite-based pollution data and demographic health surveys is that these data are comparable across countries, which allows for obtaining a more detailed picture of the expected effects of pollution across Latin America. Moreover, consideration should be given to additional outcomes, such as whether babies are premature, as well as to long-term outcomes like measures of education.

Ultimately, however, our suggestive evidence needs to be accompanied by future research focusing on additional causal inference analysis of this impact of pollution, expanding from the associations we uncover to actual impact measures. We improved upon the correlations commonly presented in the epidemiology literature by controlling for mother/family characteristics that remain constant over time and for local trends that may have an effect on weight. However, there may still be a concern that other elements that affect weight are also changing with the pollution levels we measure. In particular, changes in the level of pollution over time may be correlated with changes in economic activity and labor market outcomes, affecting families beyond the presence of suspended particles in the air. Future papers need to explore natural experiments or instrumental variables to address this concern. The use of thermal inversion as an instrument for pollution seems quite promising.24

Footnotes

  1. 1.

    For example, several cities have implemented integrated air quality management plans and have made sectoral investments, such as sustainable urban transport.

  2. 2.

    Making choices to avoid pollution depending on the gender of their future child could be another channel. This type of behavioral response can be plausible in some contexts, but corroborating evidence does not seem widespread.

  3. 3.

    Even in the epidemiology literature, evidence for developing countries, and for Latin America in particular, is limited. A recent meta-analysis of 1628 studies in the region found that most of the evidence is concentrated in a few cities (Fajersztajn et al. 2017).

  4. 4.

    The evidence in the economics literature for Latin America is even scarcer. We are aware of only a few papers studying pollution exposure in utero: for Mexico (Arceo et al. 2015), Chile (Miller and Ruiz-Tagle 2018), and Uruguay (Balsa et al. 2014). There are also few economic studies for Latin America that focus on the effects of pollution later in life (see for instance, Miller and Vela 2013).

  5. 5.

    For example, Almond et al. (2009) and Black et al. (2013) study nuclear disasters in Ukraine and Norway, respectively. Nilsson (2009) investigates the long-term impact of banning leaded gasoline in Sweden during the 1970s. Sanders (2012) studies reductions in U.S. pollution caused by the recession of the early 1980s. Isen et al. (2017) examine the U.S. Clean Air Act of the 1970s and use restricted access data on adult earnings by county and date of birth.

  6. 6.

    Other satellite measures of aerosol are also available. Total Ozone Mapping Spectrometer (TOMS) data are available since the 1970s but has been discontinued. Ozone Monitoring Instrument (OMI) data are more accurate than TOMS, but start in 2005. However, MODIS has high spectral resolution, which enables it to detect clouds and aerosols better than previous satellite-based instruments.

  7. 7.

    Also, some ground stations have the advantage of monitoring additional pollutants (i.e., ozone, sulfur, and nitrogen oxides), but this is often not the case in many developing countries.

  8. 8.

    Further, while ground-based stations measure only dry particles, satellite-based measures cannot distinguish water vapor from other particles.

  9. 9.

    The name of the aerosol data used is MODIS/Terra Aerosol Optical Thickness Daily L3 Global 0.05Deg CMA (MOD09CMA). Other MODIS products include aerosol measures over water based on MODIS Aqua satellite.

  10. 10.
  11. 11.

    The MOD09CMA data can take values slightly higher than 5000 because of the processing.

  12. 12.
  13. 13.

    Particulate matter can be of different sizes, but the most common particle sizes are PM10 and the finer PM2.5.

  14. 14.

    A more detailed validation of AOD, also relying on WHO data, is presented in Gendron-Carrier et al. (2018). Based on several linear models, those authors show that AOD tracks particulate matter measures relatively closely even in a simple model without additional controls. In general, they obtain R-squares of over 0.75. Though their analysis can provide a basis for translating the AOD measure into PM10, it does not fit our data closely, probably because we use different versions of MODIS AOD data (the L3 data used here have been processed further for use in climate modeling). Instead, we rely on a coarser rule of thumb to translate AOD to PM10 measures. In Fig. 2, the left axis shows the raw AOD measure, whereas the right axis shows the AOD transformation to PM10. The two axes can be used to coarsely translate between both pollution measures. For reference, the horizontal line in the figure gives WHO’s recommended maximum annual average PM10 exposure level of 20 mg/m3.

  15. 15.

    Not all of this variation is used in this paper, as the DHS data are not available for all municipalities/districts and all time periods.

  16. 16.

    To construct the figure, we first calculated monthly pollution averages in each country so that the box plot captures only time variation in pollution and not geographic variation.

  17. 17.

    Median is chosen to better capture first moment of distribution in a context in which frequency of observation may be low.

  18. 18.

    The following steps were used in preparing the maps: first, each child in the birth records of the DHS was assigned the pollution level in the municipality where the mother resided during pregnancy; then, after each child’s pollution exposure was determined, we calculated the average municipal exposure for children in the survey.

  19. 19.

    One caveat of this decision is that we implicitly select the sample by relative unresponsiveness to pollution shocks in terms of geographic mobility.

  20. 20.

    Consequently, the expected change in the health measure for girls and boys would be equal to (0.01 ∗ β1 ∗ δ)) and (0.01 ∗ (β1 + β3) ∗ δ), respectively.

  21. 21.

    When interpreting the results in Table 1, it is important to take into account that the satellite-based pollution measures used in the regressions should be interpreted as logarithms. Given that the model is linear-log, for a 1% change in pollution, the effect is calculated as the coefficient on pollution divided by 100. For a 10% change in pollution, the effect is calculated as the coefficient on pollution divided by 10.

  22. 22.

    In all models, the male dummy is positive and significant, suggesting that males have a higher birth weight by approximately 150 to 200 g.

  23. 23.

    To check for robustness of the results and comparability across samples, we run the estimations from columns (1), (2), and (3) using the restriction of families that have more than one child. Results are comparable in magnitudes but sometimes not statistically significant.

  24. 24.

    Temperature inversions are climatologically phenomena that are independent of short-run variations in local pollution as well as local economic activity.

Notes

References

  1. Almond D, Edlund L, Palme R. Chernobyl’s subclinical legacy: prenatal exposure to radioactive fallout and school outcomes in Sweden. Q J Econ. 2009;44.Google Scholar
  2. Arceo E, Hanna R, Oliva P. Does the effect of pollution on infant mortality differ between developing and developed countries? Evidence from Mexico City. Econ J. 2015;126:257–80.  https://doi.org/10.1111/ecoj.12273.CrossRefGoogle Scholar
  3. Bacci S, Bartolucci F, Chiavarini M, Minelli L, Pieroni L. Differences in birthweight outcomes: a longitudinal study based on siblings. Int J Environ Res Public Health. 2014;11:6472–84.  https://doi.org/10.3390/ijerph110606472.CrossRefGoogle Scholar
  4. Balsa AI, Bloomfield J, Caffera M. The effect of acute and intensive exposure to particulate matter on birth outcomes in Montevideo. IDB working paper no. IDB-WP-534. Washington, DC: Inter-American Development Bank; 2014.Google Scholar
  5. Bharadwaj P, Zivin JG, Gibson M, Neilson C. Gray matters: fetal pollution exposure and human capital formation. J Assoc Environ Resour Econ. 2017;4:505–42.Google Scholar
  6. Black S, Bütikofer A, Devereux P, Salvanes K. This is only a test? Long-run impacts of prenatal exposure to radioactive fallout. NBER working paper no. w18987. Cambridge: National Bureau of Economic Research; 2013.  https://doi.org/10.3386/w18987.CrossRefGoogle Scholar
  7. Cifuentes LA, Krupnick AJ, O’Ryan R, Toman MA. Urban air quality and human health in Latin America and the Caribbean. Washington, DC: Inter-American Development Bank; 2005.Google Scholar
  8. Clougherty JE. A growing role for gender analysis in air pollution epidemiology. Environ Health Perspect. 2011;118(2):167–76.CrossRefGoogle Scholar
  9. Currie J, Neidell M, Schmieder JF. Air pollution and infant health: lessons from New Jersey. J Health Econ. 2009;28:688–703.  https://doi.org/10.1016/j.jhealeco.2009.02.001.CrossRefGoogle Scholar
  10. Currie J, Zivin JG, Mullins J, Neidell M. What do we know about short- and long-term effects of early-life exposure to pollution? Ann Rev Resour Econ. 2014;6:217–47.  https://doi.org/10.1146/annurev-resource-100913-012610.CrossRefGoogle Scholar
  11. De Zegher F, Francois I, Boehmer AL, Saggese G, Müller J, Hiort O, et al. Androgens and fetal growth. Horm Res Paediatr. 1998;50:243–4.CrossRefGoogle Scholar
  12. De Zegher F, Devlieger H, Eeckels R. Fetal growth: boys before girls. Horm Res Paediatr. 1999;51:258–9.CrossRefGoogle Scholar
  13. Drevenstedt GL, Crimmins EM, Vasunilashorn S, Finch CE. The rise and fall of excess male infant mortality. Proc Natl Acad Sci. 2008;105:5016–21.  https://doi.org/10.1073/pnas.0800221105.CrossRefGoogle Scholar
  14. Fajersztajn L, Saldiva P, Amador Pereira LA, Figueiredo Leite V, Buehler AM. Short-term effects of fine particulate matter pollution on daily health events in Latin America: a systematic review and meta-analysis. Int J Public Health. 2017;62:729–38.CrossRefGoogle Scholar
  15. Foster A, Gutierrez E, Kumar N. Voluntary compliance, pollution levels, and infant mortality in Mexico. Am Econ Rev. 2009;99:191–7.  https://doi.org/10.1257/aer.99.2.191.CrossRefGoogle Scholar
  16. Gendron-Carrier N, Gonzalez-Navarro M, Polloni S, Turner M. Subways and urban air pollution. NBER working paper no. w24183. Cambridge: National Bureau of Economic Research; 2018.  https://doi.org/10.3386/w24183.CrossRefGoogle Scholar
  17. Hansen-Lewis J. Does air pollution lower productivity? Evidence from manufacturing in India. Brown University; 2018. Available at: http://barrett.dyson.cornell.edu/NEUDC/paper_324.pdf.
  18. Isen A, Rossin-Slater M, Walker WR. Every breath you take—every dollar you’ll make: the long-term consequences of the Clean Air Act of 1970. J Polit Econ. 2017;125:848–902.CrossRefGoogle Scholar
  19. Jayachandran S. Air quality and early-life mortality: evidence from Indonesia’s wildfires. J Hum Resour. 2009;44:916–54.  https://doi.org/10.3368/jhr.44.4.916.CrossRefGoogle Scholar
  20. Jedrychowski W, Perera F, Mrozek-Budzyn D, Mroz E, Flak E, Spengler JD, et al. Gender differences in fetal growth of newborns exposed prenatally to Sirborne fine particulate matter. Environ Res. 2009;109(4):447–56.CrossRefGoogle Scholar
  21. Lampl M, Gotsch F, Kusanovic JP, Gomez R, Nien JK, Frongillo EA, et al. Sex differences in fetal growth responses to maternal height and weight. Am J Hum Biol. 2009;22:431–43.  https://doi.org/10.1002/ajhb.21014.CrossRefGoogle Scholar
  22. Miller S, Ruiz-Tagle JC. Adverse effects of air pollution on the probability of stillbirth delivery: evidence from Central Chile. IDB discussion paper no. IDB-DP-00616. Washington, DC: Inter-American Development Bank; 2018.Google Scholar
  23. Miller S, Vela MA. The effects of air pollution on educational outcomes: evidence from Chile. In: IDB working paper no. IDB-WP-468. Washington, DC: Inter-American Development Bank; 2013.  https://doi.org/10.2139/ssrn.2370257.CrossRefGoogle Scholar
  24. Molina T. Pollution, ability, and gender-specific investment responses to shocks. Job Market Paper; 2016. Available at: https://dornsife.usc.edu/assets/sites/524/docs/Seminars_Fall_2016/Molina_Pollution_Sep2016.pdf.
  25. Nilsson JP. The long term effects of early childhood Lead exposure: evidence from the phase-out of leaded gasoline. Uppsala University. Unpublished; 2009.Google Scholar
  26. Provençal S, Kishcha P, da Silva AM, Elhacham E, Alpert P. AOD distributions and trends of major aerosol species over a selection of the world’s most populated cities based on the 1st version of NASA’s MERRA aerosol reanalysis. Urban Clim. 2017;20:168–91.  https://doi.org/10.1016/j.uclim.2017.04.001.CrossRefGoogle Scholar
  27. Sanders NJ. What doesn’t kill you makes you weaker: prenatal pollution exposure and educational outcomes. J Hum Resour. 2012;47:826–50.  https://doi.org/10.1353/jhr.2012.0018.CrossRefGoogle Scholar
  28. Sanders NJ, Stoecker C. Where have all the young men gone? Using sex ratios to measure fetal death rates. J Health Econ. 2015;41:30–45.  https://doi.org/10.1016/j.jhealeco.2014.12.005.CrossRefGoogle Scholar
  29. UNICEF. Clear the air for the children: the impact of air pollution on children. New York: UNICEF; 2016.Google Scholar
  30. Van Vliet G, Liu S, Kramer MS. Decreasing sex difference in birth weight. Epidemiology. 2009;20:622.  https://doi.org/10.1097/EDE.0b013e3181a82806.CrossRefGoogle Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  • Gabriela Aparicio
    • 1
  • María Paula Gerardino
    • 2
    Email author
  • Marcos A. Rangel
    • 3
  1. 1.IDB InvestWashingtonUSA
  2. 2.Inter-American Development BankWashingtonUSA
  3. 3.Duke UniversityDurhamUSA

Personalised recommendations