Early-life exposure to air pollution: Effect on child health in India

  • Blog Post Date 02 January, 2020
  • Print Page
Author Image

Kunal Bali

Indian Institute of Technology Delhi


Author Image

Sourangsu Chowdhury

Indian Institute of Technology Delhi


Author Image

Sagnik Dey

Indian Institute of Technology Delhi


Author Image

Prachi Singh

Indian Statistical Institute, Delhi Centre


More than half of Indian population gets exposed to PM2·5 greater than the annual limit recommended by the National Ambient Air Quality Standards. This article examines the effect of outdoor air pollution on child health by combining satellite PM2.5 data with geo-coded Demographic and Health Survey of India. It finds that children exposed to high levels of pollution in their early lives have worse child health outcomes than those exposed to lower levels of pollution.

Exposure to harmful levels of air pollution causes respiratory problems in both adults and young children (Chakravarti et al. 2019, Neidell 2004), while in case of infants, in-utero (in the womb) exposure to air pollution has been linked to mortality (Currie and Walker 2011). The in-utero period is of special significance as it critically determines mortality outcomes, disease prevalence, and future health outcomes, abilities, and earnings. At this critical stage, fetal growth if restricted, can negatively affect future outcomes. Expecting mothers who get exposed to air pollution can suffer from respiratory distress, and any resulting inflammation in the mother’s body can potentially harm fetal growth.

In our recent work (Singh et al. 2019), we examine the effect of exposure to local air pollution in early life (the in-utero period) on child growth factors that include height-for-age (stunting measure) and weight-for-age (underweight measure). The literature linking air pollution to child health has mostly focused on child mortality. Some studies exploit natural experiments like change in air pollution regulatory or certification policies to causally infer its effect on infant mortality (Greenstone and Hanna 2014, Foster et al. 2009). On the other hand, few recent studies have focused on pollution-enhancing activities like biomass burning to assess its impact on a host of child health-related outcomes like birth weight, gestational age, infant mortality, and adult height (Rangel and Vogl 2018, Soo and Pattanayak 2019, Pullabhotla 2018). We add to this literature by examining the effect of air pollution on child's growth indicators conditional on child's survival.

Combining survey data with satellite data

There is paucity of good-quality ground monitor-based air pollution data, which covers all parts of India. This is mainly due to the fact that there are less than 600 such monitors to cover the entire nation with almost no coverage for rural areas. To address the paucity in ground-based pollution data in India, we estimate fine particulate matter (PM2.5) exposure using satellite data (van Donkelaar et al. 2010, Dey et al. 2012). In Figure 1, we plot mean pollution (average over 2010 to 2016) at the district level. The figure shows that northern states have higher levels of pollution in comparison to southern states. Further, within northern states, the states which lie in the Indo-Gangetic plains record the highest levels of pollution as shown by a darker shade.

Figure 1. Mean pollution (PM2.5) in India across districts of India: 2010-2016

Source: Satellite data on pollution processed by co-authors of this study.

This data on PM2.5 is combined with a health survey data for India (National Family and Health Survey (NFHS)–IV). The clusters (of households) which are sampled in this survey are geo-coded, which allows us to construct local1 measures of exposure to pollution for children who belong to these clusters. We use the cluster location, date of birth, and pregnancy duration for each child to create trimester level (3-month periods) mean pollution exposure for each child in our sample. The NFHS-IV captures measures of human body such as height-for-age (HFA-Z) and weight-for-age (WFA-Z) for children aged below five years which are our outcomes of interest. We plot the relationship between child growth indicators and age in Figure 2 for two groups of children – one with low level of exposure to pollution (1st Quintile) versus those who had high level of exposure (5th Quintile) to pollution during their first trimester. We find that children who are exposed to high levels of pollution have worse child health outcomes as shown by the solid plot that lies below the dashed plot for children who were exposed to lower levels of pollution.

Figure 2. Relationship between child growth indicators and age for children with different levels of exposure to pollution in first trimester

Note: Polynomial fit plot between height-for-age and child's age in months for children who had low level of exposure to pollution (1st Quintile) versus those who had high level of exposure (5th Quintile) to pollution during their first trimester. Shaded area is 95% confidence interval2.

Our estimation sample comprises of close to 180,000 children for whom complete pollution exposure history was available for the in-utero period. We use an instrumental variable (IV) strategy3 to identify the effect of air pollution on child health. Using wind direction, we are able to identify upwind fire events4 (exogenous or external variable) in neighbouring areas, which we use as an instrument for local pollution levels (endogenous or variable determined internally in the model)5. We also control other demographic characteristics of the household, mother, and child.

Effect of exposure to pollution on child growth indicators

Our analysis shows that fire events affect pollution positively (the first stage of an IV regression). One standard deviation6 change in upwind fire-events is associated with an increase in PM2.5 by 0.105 standard deviation units or 3.35 ug/m3. Our main results show that exposure to air pollution during the first trimester has a negative effect on child growth indicators. A standard deviation unit change in mean PM2.5 during first trimester leads to a decrease in weight-for-age z-score (WFA-Z, underweight measure) by -0.102 standard deviation units and decreases height-for-age z-score (HFA-Z, stunting measure) by -0.115 standard deviation units, which translates into a 6.7% decrease in WFA-Z and 7.8% decrease in HFA-Z. We also conduct various robustness tests to establish the validity of our results.

Effect on GDP

Stunting affects GDP (gross domestic product) of a nation via three channels: lower returns to lower education, lower returns to lower height, and lower returns to lower cognition. For India, where 66% of the workforce was stunted in childhood, a study estimates that a complete elimination of stunting would have increased GDP by 10% (Galasso et al. 2016). We use an estimate of probability of being stunted due to exposure to outdoor pollution, and find that one standard deviation increase in outdoor pollution leads to a 0.18% reduction in GDP.

Identifying vulnerable sections of the population

We take into account the variation in our data to further assess if a section of the population is more likely to suffer from the negative effects of pollution. We split our sample into poor and rich samples using wealth index of a household, and find that the negative effect of pollution on child health is present only for poor households. This can possibly be due to the fact that children in poor households have less access to healthcare to abate harmful effects of pollution on health. We also find that the negative effect of pollution on child health is limited to northern states, which have a much higher level of pollution in comparison to southern states.

India needs effective policies regarding regulation and management of outdoor pollution. Various policies and initiatives directed at reducing major sources of outdoor air pollution, related to transport, biomass burning, and coal combustion have been ineffective in the past. For example, for effective management of forest fires, the central government has a dedicated budget allocation; however, this allocated amount is really small and remains unused in every financial year. Similarly, to reduce crop residue burning activities by farmers, the government has committed itself to subsidising the use of Happy-Seeder technology; however, the uptake of this policy remains quite low due to high initial investment in the machine. Further, legal bans on crop burning have not had any ground-level impact in the past. The National Clean Air Programme (2018) is a welcome step in this domain as it plans to extend the air quality monitoring network, conduct intensive awareness and monitoring campaigns, and create city-specific action plans, among other initiatives to reduce air pollution levels.


  1. Local measure of pollution is defined as mean PM2.5 in the 75km radius around the cluster location.
  2. A 95% confidence interval is a range of values; there is a 95% probability that this range of values contains the true mean of the population.
  3. Instrumental variables are used in the regression analysis when there is the problem of endogeneity. It happens when the outcome and predictor of interest are determined simultaneously or when both are correlated with an omitted variable in the model.
  4. Data on fire events (biomass burning events or crop burning or forest fires) comes from Fire Information for Resource Management System. This data provides pixel-level information (1km*1km resolution) on fire-events. We tag each fire-event with a wind direction and use the location of a sampled cluster to determine whether a fire event is an upwind (wind blowing away from a fire event towards a cluster) or a downwind fire event (not affected by smoke from a fire event as the wind blows in a different direction).
  5. Local pollution levels are endogenous in an empirical exercise that links child health to local pollution levels. Household income and behavioural choices about use of dirty fuels for cooking or engaging in pollution enhancing activities like crop burning are omitted variables, which make estimates of local pollution levels on child health biased, and hence, unreliable.
  6. Standard deviation is a measure that is used to quantify the amount of variation or dispersion of a set of values from the mean value (average) of that set.

Further Reading

No comments yet
Join the conversation
Captcha Captcha Reload

Comments will be held for moderation. Your contact information will not be made public.

Related content

Sign up to our newsletter