Temperature Shocks, Maternal Health, and the Intergenerational Economic Burden of Climate Change in India

Kotipalli Venkata Sriram · Lanka Karthikeya · under the guidance of Dr. Vikas Kumar
Ecology, Economy and Society — the INSEE Journal
Special Section: Economics of Climate Change in Developing Countries

---

Abstract

This study estimates the impact of prenatal heat stress on birth outcomes and the resulting economic burden across India. Using 1.32 million births from the National Family Health Surveys (NFHS-4 & NFHS-5, 2010–2021) merged with high-resolution IMD temperature data, we show that each additional day above 35°C during pregnancy significantly increases the probability of low birth weight (LBW). Effects are largest in the first trimester, consistent with the fetal origins hypothesis. The economically disadvantaged, rural populations, and households in the Indo-Gangetic Plain (IGP) are disproportionately affected. The central estimate of the annual economic burden is Rs. 104.98 billion (~90% CI: Rs. 73.88–141.70 billion), dominated by intergenerational productivity losses.

---

Table of Contents

• Key Findings
• Data Sources
• Repository Structure
• Empirical Strategy
• Results
• Economic Burden
• Replication
• Citation

---

Key Findings

Finding	Estimate
Effect of one additional hot day (>35°C, T3) on LBW probability	+1.25 pp (β = 0.0125, p < 0.001)
Effect per standard deviation of exposure (27 days)	+1.25 pp in LBW incidence
Excess LBW births nationally per SD increase	~312,900
First-trimester effect (strongest)	β = 0.0174***
Second-trimester effect	β = 0.0167***
Third-trimester effect (preferred spec)	β = 0.0125***
Wealth gradient: poorest (Q1) vs richest (Q5)	β = 0.0162 vs β = 0.0048
IGP states vs non-IGP	β = 0.0150 vs β = 0.0098
Central economic burden estimate (Monte Carlo)	Rs. 104.98 billion/year
90% confidence interval	Rs. 73.88 – 141.70 billion

15 out of 16 robustness specifications yield positive, significant coefficients. The lone exception (district FE) reflects a well-documented Simpson's Paradox driven by the cross-sectional nature of heat exposure in India.

---

Data Sources

1. Climate Data — India Meteorological Department (IMD)

File	Format	Description
imd_rainfall.nc	NetCDF	Daily rainfall gridded data
imd_tmax.grd	GRD	Daily maximum temperature gridded data

Source: India Meteorological Department — imdpune.gov.in
High-resolution gridded daily Tmax matched to Census 2011 district boundaries using area-weighted spatial extraction.

---

2. Birth & Health Data — National Family Health Survey (NFHS)

Survey Round	Period	Births
NFHS-4	2015–16	575,946
NFHS-5	2019–21	747,836
Combined	2010–2021	1,323,782

Source: DHS Program — dhsprogram.com

Each NFHS round includes three record types used in this study:

Record Type	Folder Code	Files
Births Recode	IABR74DT	.DCT .DO .DTA .FRQ .FRW .MAP
Household Recode	IAHR74DT	.DCT .DO .DTA .FRQ .FRW .MAP
Individual (Women's) Recode	IAIR74DT	.DCT .DO .DTA .FRQ .FRW .MAP

The analytical LBW sample covers 356,074 births with valid birth weight data (188,286 from NFHS-4; 167,788 from NFHS-5). Mortality outcomes use a broader sample of 723,444 births.

---

3. Health Infrastructure — National Health Systems (NHS) Resource Centre

File	Period
NHS2015-16_all.xls	2015–16
NHS2019-20_all.xls	2019–20

Source: National Health Mission — nhm.gov.in
District-level health infrastructure indicators: facility availability, institutional delivery rates, and capacity measures.

---

4. Healthcare Costs — National Health Systems Resource Centre (NHSRC)

File	Format	Description
NHSRC Costing Report	.xlsx	Unit costs for neonatal intensive care and delivery services

Source: NHSRC — nhsrcindia.org
Used to estimate the public health system burden component of the economic cost framework.

---

5. Household Expenditure — NSS 75th Round (2017–18)

File	Format	Description
Household Social Consumption — Health	.pdf report	Out-of-pocket healthcare expenditure by income decile

Source: MOSPI — mospi.gov.in / MicroData
Used to compute excess out-of-pocket (OOP) expenditure per LBW birth across wealth quintiles.

---

6. Spatial Data — Census District Boundaries

District Boundary Shapefiles

File	Description
India-Districts-2011Census.shp	Main polygon shapefile
India-Districts-2011Census.dbf	Attribute table
India-Districts-2011Census.prj	Projection file
India-Districts-2011Census.shx	Spatial index

Source: Datameet — github.com/datameet/maps

District Population & Crosswalk

File/Folder	Description
delimitation_new_district_order_2008.pdf	District delimitation reference
SHRUG/shrug-pc-keys-csv/	SHRUG primary census keys
SHRUG/shrug-shrid-keys-csv/	SHRUG SHRID spatial crosswalk

Source: SHRUG — devdatalab.org
The SHRUG crosswalk harmonizes NFHS-4 and NFHS-5 district identifiers against static Census 2011 boundaries, enabling consistent exposure assignment across both survey waves.

---

7. Forward Projections — ERA5 Reanalysis

File	Format	Description
era5-x0.25_timeseries_tasmax_tas_timeseries_monthly_1950-2023_mean_historical_era5_x0.25_mean.xlsx	.xlsx	Monthly mean Tmax and Tas at 0.25° resolution, 1950–2023

Source: Copernicus Climate Data Store — cds.climate.copernicus.eu
Used for (i) validation of IMD gridded temperature data and (ii) forward projection of hot-day counts under continued warming scenarios.

---

Empirical Strategy

Primary Specification

The baseline OLS model with birth-month fixed effects (M3):

LBWidt = α + β₁ · HotDays35dt + X'idt γ + γm + εidt

where:

• LBWidt = 1 if birth weight < 2,500g; child i, district d, time t
• HotDays35dt = days with Tmax > 35°C during the relevant trimester
• X'idt = maternal age, education, parity, ANC visits, wealth quintile, rural/urban, housing quality, cooking fuel
• γm = birth-month fixed effects (absorb common seasonality)
• Standard errors clustered at district level

Model Ladder (Stepwise Sensitivity)

Model	Specification	Coefficient	SE	p-value
M1	Raw OLS	0.0067	0.0006	< 0.001
M2	+ Individual controls	0.0067	0.0006	< 0.001
M3	+ Month FE (PRIMARY)	0.0125	0.0009	< 0.001
M4	+ District + Month FE	-0.0073	0.0012	< 0.001
M5	+ State + Month FE	0.0031	0.0040	0.440
M6	+ State + Year + Month FE	0.0034	0.0042	0.414

The sign reversal in M4 is a classic Simpson's Paradox: hotter districts also have better health infrastructure (referral hospitals, NICUs), so district FE absorbs the true cross-district signal. Birth-month FE (M3) is the preferred identification strategy.

Robustness to Unobservables — Oster (2019) Bounds

Setting Rmax = 1.3 × R̃² and δ = 1 (equal selection on observables and unobservables), the bias-adjusted β* remains strictly positive. The δ required to drive β* to zero exceeds 1 substantially, meaning unobservables would need to be far more influential than wealth, education, and housing combined — unlikely in this setting.

---

Results

Main Effects (Table 2)

Specification	Coefficient	SE	N
Hot Days > 35°C, T3 [PRIMARY M3]	0.0125*	0.0009	356,074
Hot Days > 33°C, T3	0.0098***	0.0008	356,074
T3 + Rainfall control	0.0077***	0.0009	356,074
T3 + ERA5 validation	0.0127***	0.0009	356,074
Hot Days > 35°C, T1	0.0174***	0.0009	356,074
Hot Days > 35°C, T2	0.0167***	0.0009	356,074
Neonatal mortality	0.0023***	0.0003	723,444
Infant mortality	0.0027***	0.0003	723,444

*** p < 0.01

Trimester Gradient

The effect follows a monotonically decreasing gradient from T1 to T3, consistent with the fetal origins hypothesis and the biological primacy of organogenesis in early pregnancy:

T1: β = 0.0174  >  T2: β = 0.0167  >  T3: β = 0.0125

Heterogeneity (Table 3)

Subgroup	N	LBW Rate	Coefficient	BH-corrected
Q1 — Poorest	80,190	19.15%	0.0162***	***
Q2	80,739	17.73%	0.0145***	***
Q3	74,143	16.38%	0.0141***	***
Q4	66,255	15.77%	0.0119***	***
Q5 — Richest	54,747	14.15%	0.0048*	*
Rural	271,443	17.15%	0.0125***	***
Urban	84,631	15.91%	0.0102***	***
IGP States	136,069	18.00%	0.0150***	***
Non-IGP	220,005	16.15%	0.0098***	***
Poor Housing	92,994	17.80%	0.0176***	***
Good Housing	263,080	16.52%	0.0102***	***
No Electricity	25,705	18.94%	0.0097**	**
Has Electricity	330,369	16.69%	0.0121***	***
Anaemic	115,190	17.97%	0.0125***	***
Non-Anaemic	231,832	16.26%	0.0115***	***

15/15 positive, 14/15 significant after Benjamini-Hochberg correction.

Robustness (Table 4) — 15/16 Specifications Positive

Specification	Coefficient	N
Baseline M3	0.0125***	356,074
Threshold 33°C	0.0098***	356,074
+ Rainfall	0.0077***	356,074
+ Drought flag	0.0127***	356,074
+ Anaemia control	0.0124***	347,022
+ Housing quality	0.0123***	356,074
+ ERA5 trend	0.0127***	356,074
+ Wave FE	0.0125***	356,074
Long-term residents only	0.0131***	242,363
First births only	0.0142***	99,326
Rural subsample	0.0132***	271,443
Urban subsample	0.0109***	84,631
IGP states only	0.0162***	136,069
Non-IGP states	0.0104***	220,005
District FE (M4)	-0.0073***	356,074

Machine Learning Validation

Gradient Boosting classifier trained on individual, household, and climate features:

Model	AUC
Logistic Regression	0.5684 ± 0.0026
Random Forest	0.5767 ± 0.0032
Gradient Boosting	0.5788 ± 0.0027

Climate variables account for 56.5% of feature importance in the Gradient Boosting model. The moderate AUC reflects the inherently multifactorial nature of birth outcomes, not a failure of the climate signal.

---

Economic Burden

The cost framework decomposes the total burden into three components, following Ostro (1994) and Nayak, Roy and Chowdhury:

Cost Components (Central Estimate)

Component	Estimate (Rs. Billion)	Share	Source
Household out-of-pocket (OOP)	2.66	2.4%	NSS 75th Round
Public healthcare system	3.52	3.2%	NHSRC costing studies
Intergenerational productivity loss	102.56	94.4%	Victora et al.; 12% earnings reduction, 4% discount rate
Total (Central)	108.74	100%

Monte Carlo Simulation (n = 10,000)

Parameters drawn jointly:

• Earnings reduction: Uniform[0.08, 0.15]
• Discount rate: Uniform[0.03, 0.05]
• Healthcare cost parameters: varied ±15%

Statistic	Value (Rs. Billion/year)
Median	104.98
Mean	~109
90% CI lower	73.88
90% CI upper	141.70

The total burden is equivalent to 4.5% of central government health expenditure and exceeds annual allocations to PMMVY and JSY combined. Over 94% of costs are intergenerational — invisible in short-run health accounting.

---

Replication

Requirements

pip install pandas numpy scipy statsmodels scikit-learn geopandas xarray netCDF4 matplotlib seaborn

Stata is required for NFHS .DTA preprocessing (see 02_nfhs_cleaning.do).

Run Full Pipeline

python code/fullanalysis.py

This executes all ten analysis steps sequentially:

1. Data loading and merging (1.32M births)
2. Table 1 — Descriptive statistics
3. Stepwise specification table (M1–M6)
4. Table 2 — Main results (11 specifications)
5. Trimester pattern
6. Table 3 — Heterogeneity with BH correction
7. Robustness checks
8. Placebo tests and identification validation
9. Oster (2019) omitted variable bounds
10. Machine learning validation
11. Economic burden Monte Carlo (n = 10,000)

Sample Output

[4] Table 2 — Main Results...
  → LBW ~ T3 hot days>35°C  PRIMARY [M3]    0.0125  0.0009  14.40  0.0000  ***  356,074
    → +1.252pp/SD | +125.2/10k | +312,898 national

[11] Economic burden...
  MC: Rs.104.98Bn  90%CI=[Rs.73.88, Rs.141.70]Bn

---

Citation

@article{sriram2024temperature,
  title   = {Temperature Shocks, Maternal Health, and the Intergenerational Economic Burden of Climate Change in India},
  author  = {Sriram, Kotipalli Venkata and Karthikeya, Lanka and Kumar, Vikas},
  journal = {Ecology, Economy and Society -- the INSEE Journal},
  year    = {2024},
  note    = {Special Section: Economics of Climate Change in Developing Countries}
}

---

Key References

Almond, D. & Currie, J. (2011). Killing me softly: The fetal origins hypothesis. Journal of Economic Perspectives. DOI: 10.1257/jep.25.3.153

Barker, D.J.P. (1990). The fetal and infant origins of adult disease. BMJ.

Heckman, J.J. (2006). Skill formation and the economics of investing in disadvantaged children. Science. DOI: 10.1126/science.1128898

Oster, E. (2019). Unobservable selection and coefficient stability: Theory and evidence. Journal of Business & Economic Statistics. DOI: 10.1080/07350015.2016.1227711

Victora, C.G. et al. (2008). Maternal and child undernutrition: consequences for adult health and human capital. The Lancet. DOI: 10.1016/S0140-6736(07)61692-4

---

Data access for NFHS microdata requires registration at the DHS Program portal. IMD gridded data is available on request from the India Meteorological Department.