Describe
Correlations, trends and income- and region-stratified comparisons show the raw pattern and where it weakens.
RESEARCH PAPER 01 · DATA LIBRARY · VERSION 1.0
Data and methods,
kept inspectable.
This page records the sources, variables, sample rules, model sequence, verification standard and public research files behind the first GHEAS study.
CORE SAMPLE6,800country-year rows
PANEL SAMPLE4,406186 countries
PRIMARY MODELP3two-way FE + controls
VERIFICATIONPassedtolerance 0.0001
01 — SOURCE REGISTER
Six World Development Indicators.
Raw API responses are retained unchanged. Records are matched by ISO3 economy code and calendar year after aggregate entries are removed.
RoleIndicatorWorld Bank codeUnit
OutcomeLife expectancy at birthSP.DYN.LE00.INYears
ExposureAnnual mean PM2.5EN.ATM.PM25.MC.M3µg/m³
ControlGDP per capita, PPPNY.GDP.PCAP.PP.KD2021 international dollars
ControlUrban population shareSP.URB.TOTL.IN.ZS%
ControlPopulation aged 65+SP.POP.65UP.TO.ZS%
ControlHealth spending per capita, PPPSH.XPD.CHEX.PP.CDInternational dollars
Source update recorded in the stored API responses: 13 July 2026. No interpolation or imputation was used.
02 — SAMPLE FORMATION
Every exclusion remains visible.
The 2023 complete-case cross-section follows a fixed rule: use the latest year with at least 150 countries observed on every pre-specified variable.
01
World Bank economies excluding aggregates
21702
After requiring PM2.5 (µg/m³)
20003
After requiring Life expectancy (years)
20004
After requiring GDP per capita PPP
18805
After requiring Urban population (%)
18806
After requiring Population age 65+ (%)
18807
After requiring Health spending per capita PPP
18403 — ANALYTICAL PROTOCOL
Models become stronger in a recorded sequence.
Interpretation depends on coefficient size, uncertainty, direction stability and sensitivity checks—not on a single p value.
Adjust
M0–M4 add income, demography, urbanization, health resources and region using HC3 robust uncertainty.
Compare within countries
P1–P5 add country and year fixed effects, time-varying controls, a one-year lag and a three-year exposure mean.
Challenge the result
Balanced-panel and pandemic-year checks test sample dependence; an independent implementation checks the estimator.
04 — REPRODUCIBILITY
Download the evidence behind the conclusion.
CSV files contain the analysis-ready records. JSON files preserve complete model output rather than only selected headline estimates.
Core country-year data200 countries · 6,800 rows · 1990–2023Download ↓Multivariable panel data186 countries · 4,406 complete rows · 2000–2023Download ↓Country and coverage listISO3, region, income group and matched-year countsDownload ↓Cross-sectional resultsM0–M4 coefficients, uncertainty and sensitivity outputsDownload ↓Panel resultsP1–P5 and pre-specified sensitivity modelsDownload ↓Independent verificationFull-dummy OLS comparison and numerical tolerancesDownload ↓