Relationship, affects
Economic output affects Human health and longevity
The stored status is what a reviewer approved. The level below it is what the overlapping data support; the two are shown side by side and only a human moves the status.
Mechanism, evidence, reviewer judgements
The edge as stored
Economic output affects Human health and longevity4 evidenceMechanism supported
Human health and longevity depends on Economic output.
Mechanism
Higher output per person funds nutrition, sanitation, clinics, vaccines and education, and the relation between income and life expectancy across countries and time (the Preston curve) is one of the most documented in demography. The curve shifts upward over time, so part of the gain in longevity comes from knowledge and public health rather than income, and the effect flattens at high income.
Evidence
- documentPreston, S. H. (1975), The changing relation between mortality and level of economic development, Population Studies 29(2)doi.orgCross-country relation between income per person and life expectancy, and its upward shift over time.checked 2026-09-07
- documentWHO, Social determinants of healthwho.intIncome, education and living conditions as determinants of health outcomes.checked 2026-09-07
- metricgdp-per-person
- metriclife-expectancy
- Status
- mechanism_supported
- Reviewer confidence
- 0.80
- Stated strength
- 0.50
- Lag
- 365 to 3650 days
- Version
- curated_seed_v1
- Reviewed
- 2026-09-07
Confidence and strength are curated judgements recorded with the edge, not measured quantities.
Entity pages: Economic output, Human health and longevity.
Level 3 evidence, computed 2026-09-07 22:15 UTC
Correlation
GDP per person, constant 2015 US$ against Life expectancy at birth, geography WLD, 65 overlapping years (1960 to 2024). Primary statistics on diff (first differences, because at least one series has a unit root; levels are shown for comparison).
Pearson r (diff)
0.031
p 0.8088
Spearman rho (diff)
-0.001
p 0.9908
Pearson r (levels)
0.965
two trending series correlate in levels by construction
Rolling r
0.179
min -0.627, max 0.971, 5 sign changes
- no significant association on diff (n 65)
- Granger-style test significant in one direction only (x to y)
- stored status carries a documented mechanism with cited evidence (floor at 3)
Cross-correlation scan and Granger-style tests
Lead and lag
| Direction | Tested | lag 1 p | lag 2 p | lag 3 p | lag 4 p | lag 5 p | min p | Reading |
|---|---|---|---|---|---|---|---|---|
| GDP per person, constant 2015 US$ helps predict Life expectancy at birth | yes | 0.003 | 0.000 | 0.000 | 0.000 | 0.000 | 0.0000 | predictive at lag 2 (p < 0.05) |
| Life expectancy at birth helps predict GDP per person, constant 2015 US$ | yes | 0.312 | 0.675 | 0.539 | 0.469 | 0.197 | 0.1969 | no predictive information beyond own past |
Granger-style F-tests on the primary transform: whether lagged values of one series improve a regression of the other on its own lags. Predictive information is not causation; a common driver that moves one series earlier produces the same pattern.
Reading: GDP per person, constant 2015 US$ historically leads Life expectancy at birth.
- Cointegration
- no (p 0.929)
- Changepoints
- GDP per person, co: 1, Life expectancy at: 1
- ADF p (levels)
- 0.999, 0.077
- Service
- causal-service/0.1.0
Computed 2026-09-07 22:15 UTC
Causal estimate
Treatment GDP per person, constant 2015 US$ (economic-output), outcome Life expectancy at birth (human-health), controls Population with access to electricity, Gross capital formation as share of GDP, Renewable freshwater resources per person. Method backdoor.linear_regression on diff, n 25.
Estimate (outcome units per treatment unit)
0.0007
95 percent CI
-0.0008 to 0.0023
includes zero
Refutations
3 of 3 passed
| Refutation | Effect after | p | Result | Expectation |
|---|---|---|---|---|
| placebo treatment | -0.0000 | 0.485 | passed | placebo treatment (permuted) should produce an effect near zero |
| random common cause | 0.0007 | 0.431 | passed | estimate should not move materially (p is the probability the change is noise) |
| data subset | 0.0005 | 0.457 | passed | estimate should not move materially (p is the probability the change is noise) |
Assumptions
- Backdoor adjustment: conditioning on electricity-access, gross-capital-formation-share, renewable-freshwater-per-person closes every non-causal path between gdp-per-person and life-expectancy. Unmeasured common causes would bias the estimate.
- Linear, additive effect of the treatment on the outcome across the overlapping periods; no heterogeneous or threshold effects.
- Series are first-differenced (a unit-root test rejected stationarity for at least one series); the estimate is a same-period association after adjustment, not a dynamic response.
- The direction of the arrow (treatment to outcome) comes from the stored relationship and its mechanism, not from the data.
- No reverse causation and no selection into the sample of periods.
Alternative explanations
- Reverse causation: the outcome may drive the treatment.
- An omitted common cause not in the controls moves both series.
- Shared trend or regime: both series respond to the same global shock.
- Aggregation: world-level annual data hides country-level variation and timing.
- Measurement and revision: both series are estimates that are revised over time.
Identified estimand (DoWhy)
Estimand type: nonparametric-ate
### Estimand : 1
Estimand name: backdoor
Estimand expression:
d ↪
─────────────────(E[life-expectancy|gross-capital-formation-share,electricity- ↪
d[gdp-per-person] ↪
↪
↪ access,renewable-freshwater-per-person])
↪
Estimand assumption 1, Unconfoundedness: If U→{gdp-per-person} and U→life-expectancy then P(life-expectancy|gdp-per-person,gross-capital-formation-share,electricity-access,renewable-freshwater-per-person,U) = P(life-expectancy|gdp-per-person,gross-capital-formation-share,electricity-access,renewable-freshwater-per-person)
### Estimand : 2
Estimand name: iv
No such variable(s) found!
### Estimand : 3
Estimand name: frontdoor
No such variable(s) found!- no significant association on diff (n 65)
- Granger-style test significant in one direction only (x to y)
- stored status carries a documented mechanism with cited evidence (floor at 3)
- backdoor estimate: CI includes zero
Alternative explanations for the association
What else could explain it
- Both series trend with world development; the primary statistics use first differences to limit that, not remove it.
- A common driver (income, urbanisation, technology) can move both series with different lags and produce a lead/lag pattern without any direct effect.
- World aggregates hide country composition: a change in which countries report can move a world series.
- Reverse direction: Human health and longevity may act on Economic output.
- Both series are revised by their source; the analysis uses the current vintage, not the vintage available at the time.
What each rung means
Levels
- Level 0: No association found
the overlapping data show no significant association.
- Level 1: Correlation only
statistically associated in the overlapping periods; no direction, no mechanism.
- Level 2: Predictive lead/lag
one series historically contains predictive information about the other.
- Level 3: Mechanism supported
a documented physical, economic or institutional mechanism with cited evidence.
- Level 4: Causally supported
a backdoor-adjusted estimate that survived refutation, on stated assumptions.
- Level 5: Established relation
follows from engineering, accounting or scientific structure.
Every stored run, newest first
Analysis history
- 2026-09-07 22:15 causal level 3 n 65 causal-service/0.1.0
- 2026-09-07 22:15 correlate level 3 n 65 causal-service/0.1.0