Resident patent filings from 141 countries, matched against GDP, population and energy-use data. Compare countries and regions, explore the world map, or build your own hypothetical country and see what the model predicts for it.
An econometric & machine-learning study of what predicts how many patents a country files — and which countries file more or fewer than expected.
📊
Data
WIPO/World Bank patent statistics + Our World in Data economics. A 141-country cross-section plus a 2,201-observation panel (136 countries, 2000–2021).
🧩
Methodology
OLS regression, panel fixed effects, a Negative Binomial count model, and Gradient Boosting / Random Forest with country-grouped cross-validation.
💡
Findings
Population & energy use per capita robustly predict patenting. GDP per capita's effect vanishes once fixed effects are added — the income–patents link is mostly between-country, not within-country.
01
Compare countries & regions
Add up to six countries or World Bank regions and compare them side by side. Your selection is saved in the URL — use "copy link" to share this exact comparison.
Country / regionPatentsPer millionGDP / capitaPopulationvs. model
GDP per capita vs. patents per million (log-scaled · gold diamonds are region averages · orange star is your hypothetical country, section 03)
scroll or pinch to zoom · drag to pan
02
The world map
Hover any country for its numbers; click to add it to the comparison above.
● gray = not in this dataset
03
Build a hypothetical country
Drag the sliders to set a GDP per capita, population and energy use per capita, and the actual fitted regression from this analysis predicts how many patents a country with those fundamentals would file.
$689$131,990
118K1.43B
307 kWh195,336 kWh
—
predicted resident patent filings
This uses the exact fitted coefficients from Section 4 of the accompanying report: ln(1+Patents) = −29.70 + 1.11·ln(GDPpc) + 1.21·ln(Population) + 0.43·ln(EnergyPC). It's a real extrapolation from the model, not a lookup — try setting very high income with very low population to see the model's population-dominant behavior.
04
All 141 countries
Sortable, searchable. Click any column header to sort; click a row to add that country to the comparison.
Country
Patents
Per million
GDP / capita
Population
Model residual
05
Who files the most
In absolute terms, filing activity is dominated by economic scale. Click a bar to add that country to the comparison.
Top 15 countries by resident patent applications
06
Two models, compared
An interpretable linear regression (GDP per capita, population, energy use per capita → log patents) is checked against a Gradient Boosting ensemble on the same three predictors, using 5-fold cross-validation so both scores reflect genuine out-of-sample accuracy.
The performance gap between the two models is small (ΔR² ≈ 0.05) — a sign the relationship between these fundamentals and patenting activity is close to log-linear.
07
Does it hold up within countries too?
Everything above compares different countries to each other in one year. That leaves an open question: if a single country's GDP per capita rises, does its own patenting rise too — or is the cross-country pattern just "richer countries happen to patent more," for reasons that have nothing to do with income itself? To check, we rebuilt the dataset as a real panel — 2,201 observations, 136 countries, 2000–2021 — and re-ran the regression with country and year fixed effects, which strip out everything constant about a country (legal system, geography, long-run institutions) and ask only what happens when that country's own numbers move.
Coefficient (elasticity) by specification (higher = stronger association with patent filings)
GDP per capita's coefficient falls from 0.51 to 0.02 (statistically indistinguishable from zero) once fixed effects are added — the well-known link between income and patenting looks like a between-country pattern, not evidence that a country patents more as it gets richer. Population and energy use per capita, by contrast, stay positive and significant in every specification, including a Negative Binomial model fit directly on patent counts (the standard approach in the patents-economics literature since Hausman, Hall & Griliches, 1984). Full regression tables and diagnostics are in the accompanying report.
08
Over- and under-performers
Subtracting each country's model-predicted filing rate from its actual rate flags who is punching above — or below — its economic weight class. Click a bar to add it to the comparison above.
Actual vs. predicted log(patents), ranked by residual