Millions of children die before their first birthday each year, with the majority of those deaths concentrated in the developing world. The effort to reduce infant mortality is part of the United Nations’ Millennium Development Goals and has implications for long-term physical, social, and human capital. Air pollution concentrations are often significantly higher in rapidly industrializing countries than in developed countries, translating to many lives lost. Prior evidence of the causal relationship between air pollution and infant mortality has only employed data from developed countries, limiting relevance in developing world settings. Here, pollution levels and infant mortality rates are higher, preexisting vulnerabilities are more prevalent, and the cost of avoidance behavior (minimizing risk from pollution) is much more variable, depending on access to healthcare, breathing masks, good quality housing, and other mechanisms. Further, data shortages on infant births, infant deaths and pollution combined with the problem of endogeneity bias (i.e. causality between the independent and dependent variables) limit opportunities for analysis. In this study, the researchers seek to address these problems by using an instrumental variable approach to estimating the impact of air pollution on infant mortality in Mexico City in comparison with the United States.
Mexico City is a highly relevant case study because it experiences high levels of pollution and mortality common to many developing countries. At the same time, Mexico City’s pollution levels span a large range of values, the higher end of which overlaps with levels observed in the United States. The characteristics of Mexico City’s dataset allowed the researchers to estimate not only the marginal effect of pollution at a range typical of developing countries, but also to compare estimates in Mexico City to similar pollution ranges in the United States.
The researchers examined two air pollutants: PM10 and CO. PM10 is particulate matter with diameters of 10 micrometers or less, measured in micrograms per cubic meter (mg/m3), and CO is carbon monoxide, measured in parts per billion (ppb). Both pollutants are byproducts of combustion processes and are found in vehicle and industrial emissions. Two other pollutants – sulfur dioxide and ozone – were initially studied but not pursued further as they did not have a significant relationship with the instrumental variable upon which the analysis relies.
To estimate the effect of air pollution on infant mortality rates, the researchers constructed weekly, municipality-level measures of air pollution and mortality for 48 municipalities across Greater Mexico City between 1997 and2006. Then, they estimated the effect of air pollution, instrumented by thermal inversions to avoid endogeneity bias, on infant mortality (children 1 year and younger) and neonatal mortality (children 28 days and younger). The mortality effect was explored separately by whether the cause of death was likely to be pollution-related (such as respiratory and cardiovascular diseases) or not. They further compared effect estimates from Mexico City to those derived in the United States.
The researchers used a panel dataset on pollution in Mexico City, with hourly data on PM10 drawn from 10 stations from 1997 to1999 and from 16 stations starting in 2000 with hourly data on CO drawn from 24 stations. Weekly PM10 levels were constructed by averaging the maximum daily 24-hour average for PM10. Weekly CO levels were an average of the maximum daily 8-hour averages for CO. Though data were available for 56 of Greater Mexico City’s municipalities, only 48 were used as their location was within 15 kilometers of a monitoring station, a criterion that was used to increase the precision of the pollution measure. The instrumental variable (IV) approach relied on thermal inversions, wherein a mass of hot air is caught above a mass of cold air, trapping pollutants. Conditional on temperature, thermal inversions do not represent a health risk by themselves other than through the accumulation of pollutants at the ground level. The researchers proposed that the number of inversions in a given week is an instrument for pollution levels; indeed, they found that each additional inversion led to a 3.5% increase in PM10 and a 5.4% increase in CO concentrations, making thermal inversions a valid IV for those pollutants.
Results and Policy Implications
Analysis revealed that weekly PM10 pollution increases of 1 mg/m3 and 1 ppb for CO, resulted in an additional 0.24 and 0.0032 infant deaths per 100,000 live births, respectively. While the researchers found no significant effect of pollution on neonatal mortality rates aggregated by cause of death, they did find significant effects on neonatal and infant mortality rates from respiratory and cardiovascular disease and no effects on mortality from other causes.
The results demonstrate a larger marginal effect of CO on infant mortality in Mexico City than in the US, suggesting non-linearity in the effect of CO. Meanwhile, the researchers found no difference in the marginal effect of PM10 on infant mortality across geographies. These results are consistent with their findings from the Mexico City dataset alone, which suggests non-linearity in the CO effect and no evidence of non-linearity in the PM10.
The results advocate for investment in air quality improvements as one strategy to curtail infant mortality rates, particularly in urban and industrial areas that experience higher levels of harmful pollutants.
 United Nations. “Goal 4: Reduce Child Mortality.” United Nations Millennium Development Goals. http://www.un.org/millenniumgoals/childhealth.shtml
Photo credit: Gaurav Chawla via Flickr