GLOBAL HEALTH ATLAS → GLOBAL HEALTH INTELLIGENCE / V2

Money matters.
It isn’t everything.

Why do some countries live longer than others with similar economic resources?

Explore a 179-country panel, an interpretable economic baseline and the differences it leaves unexplained.

Evidence with limits. These results use a processed, publisher-imputed dataset. Associations are not causal effects. Temporal validation cannot exclude future information introduced by upstream imputation. Read the limitations.

Observation year Highlight a country

Loading verified analysis…

01 / A WORLD DIVIDED BY YEARS

The gap has a geography.

Life expectancy differs across countries and across time. National aggregates describe a country-level pattern; they do not describe every person’s experience.

Life expectancy by country

40 years85 yearsNo matched observation

Natural Earth boundaries via world-atlas. Original dataset names are reconciled explicitly; unmapped labels remain available in the country selector and table. Scroll/pinch to zoom; double-click to reset.

Country median and 10th–90th percentile life-expectancy distribution over 2000–2015 — Each country has equal weight. Bands describe the spread between countries, not confidence intervals or a world population average.

02 / MONEY MATTERS

Resources explain part of the story.

GDP is strongly skewed. A logarithmic axis exposes variation between lower-income countries while preserving the full range.

GDP vs life expectancy explorer

GDP per capita in current USD, log scale. Line: full-sample log GDP + year-indicator reference. Click a country, use keyboard focus or choose it above. This descriptive line is separate from the held-out prediction models.

03 / MONEY IS NOT EVERYTHING

What the baseline misses.

Observed life expectancy − GDP-reference life expectancy = gap

A positive gap means higher life expectancy than this economic baseline predicts. It does not establish healthcare efficiency, policy effectiveness or government performance.

Life-expectancy gap map

−20 years+20 yearsNo matched observation

Brown = below the reference; green = above. Values outside ±20 are clipped in colour only, with exact values in the selected-country readout and table. Baseline includes year indicators and is fitted to all years.

Six largest positive and negative mean country residuals against the GDP baseline, across 2000–2015 — Mean over all 16 observations per country. The complete mean/median/persistence/volatility table can be downloaded below.

Country profile

ObservedGDP reference

Inspect selected-year country data


Country	GDP / USD	Observed / years	Reference / years	Gap / years

04 / WHAT SEPARATES THE OUTLIERS?

Association needs context.

Relationships at one year, across repeated country-years, and between long-run country averages answer different questions. Switch context to see how a coefficient changes.

Analytical context Compare with life expectancy

Association explorer

Pearson and Spearman are descriptive coefficients. Repeated rows are not independent observations for inferential tests. Mortality/HIV fields are excluded from primary prediction; no intervention effect is claimed.

Conditional associations: country-clustered 95% confidence intervals
Field	Coefficient	95% CI

Model conditions on log GDP, schooling, BMI, Polio, DTP3, HepB3 and year indicators. No country fixed effects. Polio/DTP3 VIFs exceed 11, so vaccine-specific coefficients are unstable. These are ecological associations, not causal effects.

05 / CAN WE PREDICT LIFE EXPECTANCY?

Hold out the future.

Train on 2000–2010, select the model family on 2011–2012, refit on 2000–2012, and evaluate on 2013–2015. No random row split.

Mean, linear, ridge, random forest and gradient boosting MAE on 2013–2015 — Selected family is determined by validation MAE, before viewing the test metrics. The model uses ten contemporaneous covariates, excluding country IDs, year, mortality and HIV.

Temporal holdout — 537 rows, 179 countries
Model	MAE / years	RMSE / years	R²

Predicted versus observed life expectancy and holdout residuals

Test-set permutation importance as increase in MAE after shuffling each feature — 15 permutations per feature, fixed seed. Correlated features can share or mask importance; this is not a ranking of effective policies.

The upstream caveat remains. Publisher imputation used nearby years and regional averages. Without original observations and cell-level flags, these holdout scores may be optimistic. They are reproducible performance on this processed panel, not validated prospective forecasts.

06 / WHAT DID WE LEARN?

Evidence, with its boundaries.

Four limits to keep in view

Publisher-imputed data: possible upstream temporal/geographic leakage, with no cell-level flags.
Country aggregates hide within-country differences; correlations do not establish causation.
Baseline residuals depend on model form and omitted variables. Nominal GDP is not purchasing power or real income growth.
Prediction uses same-year covariates and known countries. Geographic evaluation is a separate stress test; historical results do not establish current performance.

Data audit Methodology Country gap table Model evidence Data dictionary Download project source

About the project

Global Health Atlas combines interactive D3 country exploration with reproducible Python/SQL analysis, economic-reference modelling and predictive evaluation.

Dataset: Life Expectancy (WHO) Fixed by lashagoch; publisher metadata: CC0. All 2,864 numerical rows and regional labels match the publisher download; 22 country labels differ. Publisher sources include WHO, World Bank and Our World in Data. Original source extractions were not independently replayed.

D3 (ISC), topojson-client (ISC), world-atlas (ISC) and Natural Earth public-domain geography. Dependency credits.

Money matters.
It isn’t everything.