- Data are drawn from official World Bank indicators and linked at the country-year level.
- Country metadata was used to exclude regions and broad aggregates from the analytical sample.
- The current version emphasizes transparency over precision and preserves uncertainty in interpretation.
MODULE 01 · GLOBAL COUNTRY COMPARISON · 1990–2023
PM2.5 exposure
& life expectancy
This is the first demonstration module in the wider GHEAS analysis platform. It asks what the country-level data shows, where the pattern weakens, and what must be tested before any causal interpretation is considered.
Key takeawayThe raw cross-sectional pattern is negative, but the adjusted evidence becomes much less decisive.
200 countries · 6,800 country-year observationsRELEASE 1.0 MILESTONE
A complete example of the GHEAS evidence workflow.
This view now brings together data provenance, descriptive evidence, multivariable and fixed-effects analysis, sensitivity checks, independent verification and explicit limits in one place.
01Traceable sources and processing notes
02Descriptive cross-sectional and trend evidence
03Multivariable adjustment and sensitivity summary
04Panel models, independent verification and formal paper
DATA LIBRARY · DATA PIPELINE
Traceable sources, a repeatable workflow and a clear boundary.
Source indicators
World Bank indicators EN.ATM.PM25.MC.M3 for PM2.5 exposure and SP.DYN.LE00.IN for life expectancy.
Coverage
The analysis uses 200 countries from 1990 to 2023, with matched country-year observations for the main study sample.
Processing
Country metadata was used to remove regional and income-group aggregates before the analysis sample was formed.
Boundary
This page reports descriptive and model-based associations; it does not present a causal claim.
REPRODUCIBILITY · EVIDENCE REPORT
A protocol that can be reviewed and re-run.
Country-year observations with 200 countries and 6,800 matched rows in the core sample.
Life expectancy at birth, measured in years.
Annual mean PM2.5 concentration in micrograms per cubic metre.
GDP per capita, urbanization, age 65+ share, health spending and regional indicators.
EVIDENCE SNAPSHOT
What the current data can support right now.
200 countries · 6,800 country-year rows
Negative correlation in 2023, but not decisive on its own.
The apparent association shrinks after income, demographics and health-system controls.
The present evidence supports description and comparison, not causal claims.
EXECUTIVE SUMMARY
The data show a clear descriptive pattern, but the evidence does not yet support a strong causal claim.
The simple cross-sectional comparison is negative. Once development conditions are added, the estimated relationship weakens sharply and becomes statistically inconclusive. That pattern is exactly what we would expect if national wealth and social structure are major contributors to the observed association.
AI JUDGMENT · PM2.5 E1 · CONFOUNDING E2
在加入预先确定的经济、城市化、年龄结构、医疗投入和地区变量后，PM2.5 的估计效应衰减了约 95%，置信区间跨越零。总体负相关主要受到国家发展条件影响，当前不能确认 PM2.5 与预期寿命存在独立关联。
The original negative pattern does not survive adjustment.GDP alone removes most of the estimated association. After demography, urbanization, health resources and region are included, the estimate is close to zero and its robust confidence interval crosses zero.
- Independent association
- Inconclusive · E1
- Confounding explanation
- Supported · E2
- Panel confirmation
- Not confirmed · E1
PRE-SPECIFIED ADJUSTMENT · 2023The apparent PM2.5 effect
The apparent PM2.5 effect
shrinks by 95%.
Estimates show years of life expectancy associated with a 10 µg/m³ difference in PM2.5. Horizontal lines are heteroskedasticity-robust 95% confidence intervals; crossing zero means the data do not rule out no independent association.
184 complete countriesGDP per capita PPPUrban populationAge 65+Health spendingRegion effects
MULTI-VARIABLE ANALYSIS · MODULE 02
What happens when development conditions are included?
The unadjusted estimate is negative and statistically detectable. Once income, urbanization, age structure, health resources and region are added, the PM2.5 coefficient becomes small and the confidence interval overlaps zero.
M0Unadjusted
-1.15M1+ income
-0.20M2+ demography & urbanization
+0.15M3+ health resources
+0.24M4+ region fixed effects
-0.0695% attenuation
The estimated effect falls from about -1.15 years per 10 µg/m³ to -0.06 after adjustment, leaving the adjusted estimate statistically inconclusive.
- Controls include GDP, urbanization, age structure and health spending.
- The result remains cross-sectional and ecological in nature.
- Region-level heterogeneity suggests the relationship is highly context-dependent.
SENSITIVITY CHECKS
East Asia & Pacific0.21 years
Europe & Central Asia0.33 years
Latin America & Caribbean 0.18 years
Middle East, North Africa, Afghanistan & Pakistan-0.22 years
LATEST CROSS-SECTION · 2023The aggregate pattern changes
The aggregate pattern changes
when income is held more constant.
200 countriesr = -0.279All income groups
High incomeLow incomeLower middle incomeUpper middle income
MODEL COMPARISONOne relationship,
One relationship,
four analytical views.
The sign and strength change across views. The 1990–2023 change comparison is slightly positive, while the cross-section and within-country estimates are negative. That instability is itself evidence: a single bivariate number is not a sufficient explanation.
−0.50+0.5
2023 cross-section
-0.279
All country-years
-0.278
Within-country
-0.229
1990–2023 change
+0.115
METHOD AND INTERPRETATION
How to read the evidence without overclaiming.
This first study does not yet estimate a causal effect. It compares patterns across countries and years, then tests whether the relationship remains visible after accounting for development conditions.
- The simple cross-sectional pattern is negative.
- That pattern weakens when country income and other structural factors are held more constant.
- Regional differences suggest that context matters, especially in Europe, South Asia and Sub-Saharan Africa.
North America-0.7556.1 µg/m³
South Asia-0.71240.1 µg/m³
Europe & Central Asia-0.59014.3 µg/m³
Sub-Saharan Africa -0.38728.7 µg/m³
PANEL ANALYSIS · MODULE 03 · COMPLETE
The country-year panel does not confirm an independent association.
The panel uses 4,406 observations from 186 countries during 2000–2023. Country and year fixed effects, time-varying controls, lags and sensitivity checks all point to an estimate that remains close to zero and statistically inconclusive.
+0.047 years
Two-way fixed effects plus controls; 95% CI -0.217 to +0.312.
+0.125 years
The lagged estimate remains inconclusive; 95% CI -0.143 to +0.393.
+0.141 years
The averaged exposure estimate is also inconclusive; 95% CI -0.212 to +0.493.
CONCLUSION SUMMARY
The raw association is visible, but the adjusted and panel evidence cannot confirm an independent effect.
The raw cross-sectional pattern is negative, but the relationship becomes much weaker once GDP, urbanization, age structure and health resources are included. The completed panel analysis reaches the same substantive boundary. This does not show that PM2.5 is harmless; it shows that the present ecological design cannot isolate its independent effect on life expectancy.
WHY THIS MATTERS
- It shows the value of separating descriptive patterns from causal interpretation.
- It records how year effects, controls, lags and sample checks change the estimate.
- It turns the current work into a reusable template for later evidence modules.
COUNTRY AVERAGE TREND · 1990 = 100Air pollution fell while
Air pollution fell while
life expectancy rose.
Across the matched countries, mean PM2.5 fell by 4.7 µg/m³ and mean life expectancy increased by 8.6 years. These simultaneous trends do not establish that one caused the other.
01Country-level ecological analysis cannot establish individual-level effects.
02The bivariate estimate is not adjusted for income, age structure, healthcare, smoking or other confounders.
03National means hide subnational exposure and health inequalities.
04Some source indicators are modelled estimates rather than direct measurements.
ACCEPTANCE CHECKLIST
How to verify this deliverable
01Open the homepage and confirm the project framing is clear and coherent.
02Open the analysis page and verify that the evidence story is structured from provenance to interpretation.
03Check that the adjusted estimates weaken materially and that the conclusion remains cautious.
04Confirm that the page builds locally and is ready for review, critique or extension.
DELIVERABLE STATUS
Research Paper 01 is complete as a reproducible evidence release.
The release combines data provenance, descriptive evidence, adjusted and panel models, sensitivity checks, independent numerical verification and a formal paper. It remains an observational ecological study, not a causal claim.