A machine-generated CSV mirror of the World Bank's Global Subnational Poverty Atlas (GSAP) — the deepest sub-national poverty series the World Bank publishes with global coverage.
GSAP ships only as an .xlsx file with no API, which makes it awkward to consume
from anything that speaks CSV or HTTP. This repository downloads it, flattens the
newest lineup year, and commits the result, so it can be fetched by URL.
This is not an official World Bank product. It is a reformatting of one, with no values changed beyond converting rates from fractions to percent. For the authoritative source, go to the Data Catalog entry.
data/gsap_admin1_latest.csv —
1,914 administrative units across 172 economies.
https://raw.githubusercontent.com/givedirectly/wb-poverty-data/main/data/gsap_admin1_latest.csv
| Column | Meaning |
|---|---|
iso3 |
ISO 3166-1 alpha-3 economy code |
country |
economy name from the World Bank country endpoint; falls back to the code |
admin_name |
the unit's name (sample in the source) |
admin_level |
admin1, admin2, nuts2, survey-specific, or unknown |
poor300_pct |
% of population below $3.00/day (2021 PPP) |
poor420_pct |
% below $4.20/day (2021 PPP) |
poor830_pct |
% below $8.30/day (2021 PPP) |
prosperity_gap |
average shortfall from a $28/day standard (2021 PPP) |
lineup_year |
the common year every economy's survey is projected to |
survey_year, survey_name, welfare_type |
the underlying survey (CONS = consumption, INC = income) |
geo_code, geo_code2 |
join keys for the GSAP shapefiles |
data/metadata.csv records the source URL, the published
file's ETag, the GSAP edition, and row counts for the CSV sitting beside it.
https://raw.githubusercontent.com/givedirectly/wb-poverty-data/main/data/metadata.csv
Units are administrative level 1 (state / province / governorate) for most
economies, and level 2 where the underlying survey supports it. For some
economies the units are survey-specific rather than a clean administrative tier —
admin_level says which.
Nothing finer exists on a globally comparable basis. Below GSAP there are only country-specific small-area poverty maps, published on differing methods and years.
Editions are not comparable. PPP exchange rates get revised between GSAP releases, so a unit's rate in this edition cannot be compared against a number from an earlier one. Compare within an edition only.
There are no population weights. GSAP ships rates, not populations, so these units cannot be averaged back into a national poverty rate. For national figures use the Poverty and Inequality Platform directly.
A GitHub Action checks daily and rewrites the CSV only when the published file's ETag changes — so the commit history is a log of actual World Bank releases, not of the schedule running.
Daily is worth it despite GSAP being annual: the re-cut lands unannounced
sometime in October, the underlying PIP database updates each March and September,
and a run that finds nothing new is a single HEAD request.
To rebuild by hand:
pip install -r requirements.txt
python publish.py --forcepython publish.py --check-only reports whether the source has moved without
downloading it.
The data is the World Bank's, licensed CC BY 4.0. Attribute it as:
World Bank, Global Subnational Poverty Atlas (GSAP), Poverty Global Practice. Licensed under CC BY 4.0.
The code in this repository is MIT.
Methodology is documented in On the Construction of the World Bank's Subnational Poverty and Inequality Databases, linked from the Data Catalog entry alongside the dataset itself.