Conference Agenda

I examine the causal effect of fire-induced air pollution on children’s health in India, where agricultural burning is both widespread and seasonal. Standard ap- proaches instrument air pollution exposure using wind direction and fire proximity. However, in agrarian settings like India, both fire activity and prevailing winds are highly predictable, potentially enabling behavioral avoidance and biasing causal es- timates downward. To address this, I develop a refined instrument, NetFires, which captures deviations from historical fire-wind patterns, to represent unexpected up- wind fire exposure that is less subject to anticipatory behavior.

Using geocoded data from NFHS-5 (2019–21), NASA FIRMS fire detections, ERA5 reanalysis wind fields, and high-resolution PM2.5 concentration maps, I find that one additional upwind NetFire in the final trimester leads to a 0.14-gram re- duction in birth weight (95% CI: –0.26g, –0.02g). This effect is strongest in the third trimester and is robust across alternative exposure definitions, fixed effect structures, and control specifications. Heterogeneity analysis reveals that the ad- verse effects are concentrated among rural households, poorer wealth quintiles, and female infants, highlighting environmental inequality.

Compared to conventional fire counts, the NetFires instrument yields consis- tently larger and more robust estimates, underscoring the importance of modeling exposure uncertainty. Additional robustness checks using alternative IV setups, non-upwind reclassifications, and falsification tests support the internal validity of the findings.

These results demonstrate the importance of integrating behavioral predictabil- ity into exposure modeling. NetFires offers a scalable and policy-relevant tool to identify high-risk pollution shocks in agrarian economies. The findings hold direct implications for climate-health policy, especially the design of early warning systems and protective interventions targeting pregnant women in fire-prone regions.