Maddison Project Database GDP
- Rows
- 16,143
- Columns
- 5
- Data range
- 1 to 2022
- Auto-refresh
- Re-checks its source every 30 days
Historical GDP estimates for countries and regions from Maddison Project Database.
AI analysis
Maddison Project Database GDP — Dataset Analysis
1. Dataset Overview & Value for Correlation Studies This dataset contains historical GDP estimates drawn from the Maddison Project Database, one of the most authoritative long-run macroeconomic datasets in existence, spanning centuries of economic history across countries and world regions. Sourced via Our World in Data's API at https://ourworldindata.org/grapher/gdp-maddison-project-database.csv with a version parameter (v=1) suggesting active maintenance, the dataset covers up to 2022, making it reasonably current. With 16,143 rows across 208 distinct years and 178 distinct entities (countries and aggregated regions), it offers exceptional longitudinal depth for studying long-run economic development, convergence and divergence between nations, and the macroeconomic impacts of historical events such as wars, industrialization, and globalization. Its breadth makes it a strong anchor dataset for cross-country correlation analysis.
2. Data Quality Observations Overall data quality is good but carries notable caveats. The most glaring issue is the GDP (Annotations) column, which is null in 16,122 of 16,143 rows (99.9%) — effectively a vestigial column carrying almost no informational content and safely ignorable or droppable before analysis. The Code column (ISO country codes) has 161 nulls (~1%), which likely corresponds to regional or aggregate entities (e.g., "World," "Western Europe") that lack standard ISO codes — a predictable and manageable gap. The GDP and Year columns are fully populated with zero nulls, which is excellent for time-series and regression work. No duplicate row count has been confirmed yet (pending Phase D recomputation), so deduplication should be validated before joining or aggregating. Total null cells across all columns number 16,283, but this figure is dominated almost entirely by the annotations column and does not represent systemic missingness.
3. Key Column Distributions The Year column (min=1, max=2022, mean≈1962, median=1975) reveals a strong left skew (skewness = -14.6) with 903 outliers, confirming that coverage becomes increasingly sparse the further back in history one goes — many ancient and medieval data points exist for only a handful of entities, while post-WWII coverage is dense. The GDP column is where the most analytically important distributional story unfolds: the mean of ~$404 billion dwarfs the median of ~$28.4 billion, and the skewness of 28.2 with 2,332 outliers signals extreme right-skew driven by economic giants like the USA, China, and the EU. The interquartile range ($8B–$130B) captures a wide but realistic middle band of national economies. Crucially, GDP values span nearly 7 orders of magnitude (min ~$19M to max ~$130T), making log-transformation essentially mandatory for any regression, clustering, or correlation analysis to avoid scale distortion.
4. Recommended Join Key Columns The optimal join strategy uses a composite key of Code + Year, where Code is the ISO 3-letter country code and Year is the integer year. This combination uniquely identifies country-year observations and is the standard pairing used across most international economic and demographic datasets. For regional/aggregate rows where Code is null, Entity + Year serves as the fallback join key, though string matching on entity names requires careful normalization (e.g., "United States" vs. "USA" vs. "United States of America"). Analysts should be aware that Code nulls disproportionately affect aggregate rows, so filtering to Code IS NOT NULL before joining to other country-level datasets is advisable to avoid unintended fan-out or mismatched aggregations.
5. Recommended Pairing Datasets for Correlation Discovery This dataset pairs powerfully with several complementary sources: - Population data (e.g., Maddison's own population estimates or UN Population Division) — to compute GDP per capita and study productivity and living standards over time - Trade openness / export-import volumes (World Bank WDI or OECD) — to correlate economic size with trade integration across eras - Human Development Index or literacy/education rates — to explore the relationship between human capital accumulation and long-run growth - Political regime data (e.g., V-Dem or Polity5 datasets) — to test correlations between democratic governance and economic performance historically - CO₂ emissions or energy consumption (Global Carbon Project, Our World in Data) — highly relevant given the shared source publisher, likely compatible entity naming, and the pressing question of how GDP growth correlates with environmental impact across development stages
Columns
- Year (integer)
- Code (string)
- Entity (string)
- GDP (integer)
- GDP (Annotations) (string)