Global number of reported disasters by size per decade

Rows: 1,647
Columns: 7
Data range: 1900-01-01 to 2020-01-01
Auto-refresh: Re-checks its source every 30 days
Last refreshed with new data: 2026-11-02

‘Small’ events have 5 or fewer deaths, less than 1,500 people affected, or reported economic damages below US$13 million. ‘Large’ events have at least 50 deaths, more than 150,000 people affected, or reported economic damages of at least US$320 million. At least part of the increase in small and medium reported events is due to improved reporting over time.

AI analysis

Dataset Analysis: Global Reported Disasters by Size per Decade

1. Dataset Overview and Value for Correlation Studies

This dataset, sourced from Our World in Data (ourworldindata.org/grapher/natural-disasters-by-decadal-size.csv), aggregates counts of reported natural disasters into four severity tiers—Small, Medium, Large, and Unknown—by entity (country/region) and decade, spanning 1900–2020. The "decadal" structure (only 13 distinct years, implying decade-start markers like 1900, 1910, etc.) makes this a coarse-grained but long-horizon panel dataset, well-suited for studying century-scale trends in disaster frequency and severity classification. Since OWID aggregates from EM-DAT and similar disaster databases, it's valuable for correlation work against climate indicators (temperature anomalies, CO2 emissions), economic development metrics, or population growth—though the explicit caveat that "improved reporting over time" inflates small/medium event counts means any correlation with modern variables (e.g., GDP, internet penetration) should be interpreted cautiously, as reporting-bias confounds may masquerade as real relationships.

2. Data Quality Observations

Data completeness is excellent at the cell level: zero null values across all 1,640 rows and 7 columns, and no columns show disproportionate missingness. Duplicate-row detection is still pending (marked for Phase D recomputation), so uniqueness on the (Year, Entity) composite key cannot yet be confirmed—this should be validated before using the dataset as an authoritative join source. The Entity and Code columns both show 238 distinct values, consistent with country/region-level coverage, and their alignment suggests reliable ISO-style coding rather than free-text entity naming, which is a positive sign for join integrity.

3. Key Column Distributions

All four disaster-count columns are heavily right-skewed, reflecting a classic disaster-data pattern: most country-decades report few or zero events, while a small number of high-exposure countries/decades dominate. Medium has the highest volume (mean=18.88, median=2, σ=98.41, skew=12.2, max=1880) with 200 outliers, followed by Large (mean=13.81, median=1, σ=64.51, skew=10.3, max=1091, 201 outliers) and Small (mean=5.61, median=0, σ=31.36, skew=13, max=721, 179 outliers). The Unknown category is smallest and least variable (mean=2.04, median=0, max=165, 153 outliers) but still exhibits skew=10.1, indicating classification gaps persist even at low volume. The gap between median (0–2) and mean (5.6–18.9) across all four columns confirms these are not normally distributed and require log-transformation or rank-based methods before any parametric correlation analysis (e.g., Pearson) is applied.

4. Recommended Join Keys

For cross-dataset correlation, use Year + Entity (or Year + Code) as the composite join key. Code is preferable to Entity for joining against other OWID or World Bank datasets since it likely follows standardized ISO-3 country coding, reducing string-matching errors from naming variants (e.g., "United States" vs. "USA"). Given the decadal granularity, joined datasets should either be pre-aggregated to decade level or the analyst should decide whether to broadcast decade values across all component years.

5. Suggested Complementary Datasets

Strong pairing candidates include: (a) climate datasets—global mean temperature anomalies, CO2/GHG emissions by country—to test whether disaster severity mix shifts with warming; (b) socioeconomic datasets—GDP per capita, population density, urbanization rate—to explore whether economic development correlates with damage thresholds (since "Large" is partly defined by economic loss in USD, wealthier/more populous nations may mechanically report more "Large" events); (c) governance/infrastructure indices (e.g., disaster preparedness spending, insurance penetration) to assess whether institutional capacity moderates severity classification; and (d) media/internet penetration or EM-DAT reporting-coverage metrics specifically to help disentangle genuine disaster increases from the stated reporting-bias effect over time.

Columns

Year (date)
Entity (string)
Code (string)
Small (integer)
Medium (integer)
Large (integer)
Unknown (integer)