Reproducible model evidence

GDP Regression, Significance, and Eta

The canonical model is a population-weighted cross-sectional regression of log GDP per capita on the number of activated levers. It estimates association with GDP level, not a causal time-series growth effect.

η (eta)

0.305

Implied proportional GDP-per-capita difference per additional activated lever: exp(slope)-1.

Uncertainty

8.3% to 57.3%

Approximate 95% robust interval for η.

Model fit

R² 0.252

n=187; log slope=0.266; robust SE=0.095; p=0.00568.

Exact specification

log(GDP per capita) = intercept + beta × activated-lever count + error, fitted by population-weighted least squares. η = exp(beta)-1. The atlas rounds the estimated 0.305 to 0.30, then applies ΔG = min(η × (1-R) × G₀, 0.25 × G₀).

Significance is not causality. The coefficient is statistically distinguishable from zero in this specification, but cross-sectional omitted variables, reverse causation, proxy reuse, threshold choices, and source-year mismatch remain. The 25% cap limits attribution; it does not solve identification.

Comparator checks

Single predictor	n	Slope	R²	p
Population	187	-0.161	0.063	0.000188
Undernourishment	159	-0.094	0.468	1.19e-14
Extreme-poverty headcount	61	-0.115	0.351	2.49e-05

These unweighted, one-predictor comparisons use log GDP per capita as the outcome. Undernourishment and poverty are themselves development outcomes and lever 9 inputs, so they are descriptive comparators, not independent controls. Different sample sizes reflect missing data.

Growth falsification check

n=179; slope=-0.170 percentage points per lever; R²=0.004; p=0.207.

The available 2023–2024 current-US$ change does not show a statistically significant relationship with lever count. This is why the atlas describes a strong association with the cross-sectional GDP level and does not claim that the nine levers predict one-year GDP growth.