Unequal World is a living atlas of global inequality. It joins Johnny Miller's Unequal Scenes aerial photography with authoritative statistics, a sourced library of the moments inequality changed course, narrative stories, within-city maps, and an evidence engine for checking claims. Every measured number is one click from its source, and everything we derive is labelled derived.
The platform has several parts built on the same data and the same discipline. A data atlas renders country indicators on an interactive globe across four tabs (Health, Economy, Urbanisation, Climate), with a year slider back to 1980 and a per-country dashboard. A turning-points library documents the moments national trajectories bent, every entry sourced. The story layer holds reported pieces, most anchored to a place and a few built from a global data series instead. City maps go below the national number: within-city income surfaces built from census micro-geography, and fine environmental grids. A projects layer marks verified responses to the inequality this platform maps. A claim-check engine, currently a prototype, maps the evidence behind claims about inequality without letting a language model invent any of it.
The analytics that run in the tool itself, including peer comparisons and divergences, are deterministic statistics. No model generates numbers at runtime outside the fenced claim-check prototype described below. Unequal World is made by Johnny Miller, a Senior Atlantic Fellow at the International Inequalities Institute at the London School of Economics.
The library documents the moments national trajectories bent: policy choices, collapses, recoveries and reforms, each tied to a specific country, year and indicator. Every entry carries a named, linkable source, and some cross-reference a second indicator from a different dataset, so a single moment can be read across, for example, a democracy index and a wealth share.
One library feeds every surface that shows a turning point. On the globe each entry is a pin at its country, coloured by what the change did: improvement, decline or catastrophe. The browsable library filters by theme, by kind (choices, shocks, transitions, patterns), by decade and by country. Compare puts two to four entries side by side and reads what they share. A record the audit could not confirm is held back from all three rather than shown with a caveat, and the atlas draws no pin it cannot source.
Claim check is an auditable evidence map. Give it a claim about inequality, or a link to where you heard one, and it returns a structured report: where the measured evidence sits, which named scholars support or contest the claim, the primary data behind each point, and every source. It does not issue verdicts; it maps evidence and disagreement and hands the judgement back to the reader.
It draws only on a closed, cached set of named public datasets and a hand-written file of sourced scholarly positions, never the open internet. The server discards any citation whose id does not exist in that set, checks every figure the model writes against the sourced values behind its citations, with an untraceable figure blanked or flagged to the reader, fences the scholar names it may invoke, and, on registered contested questions, automatically surfaces the side of the debate the answer omitted, with a coverage receipt. When a reader submits an outside link, the server fetches that one page and labels it external evidence; it is never merged into the measured set.
Two honest caveats. Claim check is a prototype that runs as its own firewalled service, and the door to it in the app says so when the engine is offline. And the scholarly-positions file is marked seed-for-review: researched and sourced, but not yet independently vetted, and not to be treated as endorsed until it is. The full methodology, including what the gates do and do not guarantee, ships with the engine's own documentation.
Each story is a scrolling, photographed piece about one place, or about a pattern that plays out globally. It pairs Johnny Miller's aerial photography, where the place has been photographed, with the data behind what the picture shows: a census surface, a satellite record, a published research finding. Every claim inside a story carries a source chip, so a reader can jump straight from a sentence to the series it comes from.
Find them from the Stories menu on the atlas. Open one and scroll: the charts and maps building inside the story track the numbers as you go, and the sources stay linked throughout, so nothing in the telling is unsourced. Every story links back into the wider platform: to the turning point it relates to, to the city on the globe, or to the dataset behind a global pattern.
Below the country layer, each city has its own map. A background switch shows the place as a satellite image or as a street map, and four groups of layers sit on top of it: Health (access to care, sanitation, life expectancy), Economy (household income, poverty, education, unemployment), Urbanisation (building density, building height, land developed since 2000) and Climate (air pollution, heat, flood risk, tree cover). Each layer is drawn from official figures where the city publishes them, and from satellite or global models otherwise, coloured neighbourhood by neighbourhood, so the map shows a real gap running through a single city rather than a national average that hides it.
What each city's map is built from differs by design: the economy layer might be census household income in one city and a poverty or deprivation index in another, whatever the local statistics office publishes at the finest resolution. The map always states which, and a coloured badge on the card marks how directly the layer measures what it claims to.
The city's panel lists City measures, the figures published for the city itself, and Country measures, the national series for the same country. Each row draws the full series it has, and one year slider moves them all.
A further layer combines air quality, green space and neighbourhood wealth into a single environmental score, ranked within that city only, so a reader can see which parts of a city carry the heaviest environmental burden relative to the rest of it, never compared across cities.
Two things are in progress. Scientific work is under way to estimate a fuller exposome, the whole mix of environmental exposures a resident experiences, at neighbourhood scale. And because every city currently publishes its own statistics in its own way, work continues on harmonising the data points across cities so more of them become properly comparable.
A projects layer marks verified responses to the inequality this platform maps: organisations and initiatives working in documentation and photography, arts and health, environmental and climate justice, urbanism and spatial inequality, and inequality research. Turn it on from the Projects toggle on the atlas to see them plotted alongside the data.
Each entry names its organisation, its category and its place, and links out to it. How the entries are sourced and checked is set out in the methodology below (§12).
The sections above say what Unequal World is. What follows says exactly how each value is produced, when two numbers may be compared, and where the method breaks. The rule throughout: measured data is shown as-is and source-linked, anything we derive is labelled derived, and known artefacts are flagged or suppressed rather than presented as findings.
Unequal World combines three kinds of object: (a) country indicators served from the World Bank's Data360 / WDI and its federated IMF and UNICEF series, extended with the World Inequality Database, LIS key figures, V-Dem, UNDP HDI and long-run Our World in Data series; (b) within-city surfaces we built ourselves from national census and statistics offices; and (c) satellite-derived environmental layers. It is an atlas for seeing inequality and verifying its national figures, not a statistical model of its causes.
It is not: a causal model (cross-indicator relationships shown in the country panel are correlational and labelled as such); a nowcast (every value carries the source vintage, and several within-city surfaces rest on the last available census); a clinical instrument (the environmental burden index is a population-level index, not an individual prediction); or a single-vintage, single-method dataset that can be compared cell-for-cell across all cities (see §4).
Every city carries one of three tiers, shown on its card and in its source line. The tier governs whether a value may be read as a number or only as a pattern.
Only cities measured in a real currency income, each in its own national currency, at a fine spatial unit, are flagged comparable: true and used in any cross-city ratio. A literacy rate and a household income are never placed on the same axis. Everything below the currency-income line is shown for its own internal gradient only.
Comparable here means narrow: measured in a real currency at fine resolution. It does not mean the cities measure the same construct. Each country defines household income differently (see §4), so even within this set a cross-city number is a contrast, not a like-for-like league table.
The comparable set is built from primary census micro-geography. We download the official small-area table, join median household income to the matching boundary, and render the choropleth at the finest unit the source publishes. Representative sources:
| Region / cities | Source | Spatial unit | Vintage |
|---|---|---|---|
| US (Detroit, Chicago, New York, Los Angeles and others) | US Census Bureau ACS 5-year (B19013, continuous median) | Block group | 2022 |
| Brazil (São Paulo, Rio, Brasília and others) | IBGE Censo Demográfico, setores censitários, Rendimento do Responsável (V06004: mean income of household heads who have income; no Gini derived) | Census sector | 2022 |
| South Africa (Cape Town, Johannesburg, Durban and others) | StatsSA Census, banded household income, via DataFirst (UCT) | Small Area Layer | 2011 |
| Ireland (Dublin) | CSO Geographical Profiles of Income, table GPIIA01 (Revenue + DSP administrative income) | Small area | 2022 |
The income construct is not identical across countries. US ACS is continuous pre-tax money income; the Irish figure is modelled from Revenue/DSP administrative records; the South African 2011 figure is collected in income bands, so the rand values are bracket midpoints (imputations), not measured amounts. A banded measure compresses the extremes, which directly affects any ratio drawn from it.
The within-city "Gini" is a spatial, between-area statistic. Where the platform shows a city Gini, it is computed over area medians, so it captures inequality between neighbourhoods and discards inequality within each one. It is therefore systematically lower than, and not comparable to, the national household Gini shown in the country panel. The two are different measurements that happen to share a name.
Brazilian cities show no city Gini at all. IBGE's 2022 sector income variable (V06004) is the mean income of household heads who have income, so heads without income are excluded and poor sectors read higher than they are. IBGE's own methodology note (Notas metodológicas 02/2025, p. 12) states the variable cannot support a household income Gini, poverty lines, or a ranking of places by income. Those cards therefore carry only the ratio of sector means, labelled as such.
Cape Town illustrates the resolution: its Small Area Layers carry median annual household income across a steep gradient (roughly two orders of magnitude, ratio ~16×). Because the 2011 income is banded, that ratio is sensitive to how the open-ended top bracket is imputed, so it is best read as "a very large gap," not a precise multiple. The gradient is the unit of analysis.
Two well-sourced numbers can still be incomparable. Three systematic effects are disclosed rather than hidden:
The index runs 0 to 100, higher meaning worse, and it is the one figure on the platform we compute ourselves. Each 100 m cell combines a wealth term, air quality and green space, weighted more heavily towards wealth, into a single score. Every input is ranked within its own city before it is combined, so a cell's value places it against the rest of its own city and means nothing next to a cell in another city. The city map's coloured neighbourhoods are the same grid, aggregated up to neighbourhood boundaries.
The index is editorial, not a measurement or an official statistic. The weights are our judgement, informed by the direction of effect the Legaz et al. (2026, Nature Medicine, doi:10.1038/s41591-026-04302-z) brain-ageing research found, and the picture would shift under other defensible weights. It is a population-level index, not a diagnosis of anyone living in the neighbourhoods it colours. The inputs, the weights and the source of every value are published alongside the data the site serves, so anyone can redo the arithmetic behind any cell.
Environmental layers come from ESA Sentinel-2 (vegetation / NDVI), Sentinel-5P (NO₂), NASA VIIRS (night lights) and the EU JRC GHSL (built-up and population). Satellite estimates are convenient and global but carry systematic biases we correct or flag:
Cities with a 100 m environmental grid (Santiago, Buenos Aires, Johannesburg, São Paulo, Mumbai) have a Degree of Urbanisation colouring from the EU JRC's GHS-SMOD R2023A, 2020 epoch, read through Google Earth Engine. GHSL assigns each cell one of seven classes, from very low density rural through rural cluster and suburban to urban centre, using population and built-up surface together, plus a water class. It classifies cells rather than measuring anything at them. The colours on the map are the JRC's own published colours for these classes, so the key here is the key a reader meets in every other publication that draws this dataset.
Source: Schiavina, M., Melchiorri, M., Pesaresi, M. (2023). GHS-SMOD R2023A. European Commission, Joint Research Centre. doi:10.2905/A0DF7A6F-49DE-46EA-9BDE-563437A6E2BA. The JRC catalogue entry places it under the European Commission reuse notice: reuse is authorised, provided the source is acknowledged. It names no Creative Commons licence, so none is claimed here.
Every PM2.5 figure elsewhere on this platform is modelled from satellite: the city grids, the GHSL urban-centre estimate, and the World Bank's population-weighted national series all trace back to the same family of satellite-derived surfaces. The WHO Ambient Air Quality Database, version 8.0 (released 30 June 2026) is the measured counterpart, reporting annual mean concentrations from ground instruments across a wide network of settlements. Where WHO has a station in a city this atlas draws, the city map's City measures list has a row labelled PM2.5, ground stations, with the station count and the satellite estimate for the same year in its note, and the country panel's Health section puts a row labelled PM2.5, ground stations (WHO) under the modelled card. Stations measure the air at a point; the satellite surface models a whole city. Showing both is how a reader can see where the two disagree.
Source: World Health Organization (2026). WHO Ambient Air Quality Database (Version 8.0). Geneva, World Health Organization. WHO states no Creative Commons licence on this database and directs reuse questions to its air quality team, so it is reproduced here under the citation WHO itself suggests. The shipped file records the download URL, the workbook's SHA-256 and the date it was fetched.
The Climate tab has one live layer. Active fires draws NASA FIRMS thermal anomalies from the VIIRS instruments on NOAA-20 and Suomi NPP, at 375 m, read straight from NASA's Global Imagery Browse Services with no key and no intermediary. It is the only layer on this platform whose numbers change between two visits, so the panel names the UTC day drawn and how many minutes ago the tiles were read, and it says when the feed did not answer instead of drawing an empty world.
Source: NASA FIRMS via NASA GIBS. The underlying detections are US government work in the public domain with a citation request; GIBS publishes no licence identifier and asks for acknowledgement. We acknowledge the use of data and imagery from NASA's Fire Information for Resource Management System (FIRMS), part of NASA's Earth Science Data and Information System (ESDIS).
The Climate tab's Tree cover loss layer is annual Landsat-derived loss at 30 m from the University of Maryland's GLAD lab (Hansen et al. 2013), served by Global Nature Watch, formerly Global Forest Watch. The choropleth value is that year's loss as a percentage of the country's year-2000 tree cover extent, at the 30 percent canopy-density threshold Global Nature Watch uses by default. Both dataset versions are pinned in the shipped file and shown on the map: loss from gadm__tcl__iso_change v20260424, the year-2000 extent from gadm__tcl__iso_summary v20250515. Licence CC BY 4.0. Four limits come with it, in the publisher's own words:
Some countries have under 1,000 hectares of year-2000 tree cover at this threshold (Burkina Faso 130 ha, Mauritania 33 ha, Eritrea 5 ha). Their percentages would be edge-pixel noise, so they show hectares in the country panel and no colour on the map.
Every national value is served from published institutional data and carries a one-click "Verify" chip back to the authoritative series: Data360 / WDI for World Bank indicators, WID.world for income and wealth shares, and the other sources listed below. Every indicator on the country layer is a series somebody else measured and publishes, so every one of them verifies in one click. The panel also pulls federated IMF (World Economic Outlook) and UNICEF values via Data360, and World Bank GSAP subnational poverty at the first administrative level. The country panel's peer-comparison and divergence features are deterministic statistics, not machine learning: a country is flagged where its value departs from the robust median of its income-level and region peers by more than a fixed multiple of the median absolute deviation, and the peer sample size is disclosed. Turning-point markers come from the curated library described in section 3 and nowhere else: every pin is a record that passed the audit and opens its own source. A second kind of pin, in teal, marks a divergence: a country where two indicators that move together almost everywhere came apart, such as income per head doubling while life expectancy barely moved. That pin is computed from the two published series it names, states both of them, and links each one to Data360. All such relationships are correlational, not causal.
The Atkinson index. Beside the World Bank Gini, the Economy section of the country panel reports the Atkinson index at two levels of inequality aversion (ε = 0.5 and ε = 1), taken from the open tier of the LIS Key Figures. It answers a different question from the Gini: what share of total income a country could give up, if the rest were shared equally, and leave people as well off as they are now (Atkinson 1970). ε sets how much weight the measure puts on the bottom of the distribution, and the index rises with it. The card states the ε = 1 figure in a sentence and shows both values. These are LIS survey years on disposable household income, equivalised by the square root of household size, not annual World Bank figures; 54 countries have them, so the block appears on those panels only. It is a separate measurement from the Gini in the cards above, on a different income concept, and the two are not comparable digit for digit.
The scrollytelling stories are reported pieces built on a named data spine, not illustrations. Each names its sources in-story via source chips. This list is generated from the same story order the site itself reads, grouped by subject:
The 100 m environmental burden index (§5) is our own estimate of how a place's environment (air, green space, neighbourhood wealth) presses on the people in it. Its direction comes from Legaz et al. (2026, Nature Medicine), which found the environmental exposome associated with accelerated brain ageing across 34 countries. The index is developed as a research collaboration with the BrainLat Research Network, with support from the Max Planck Institute for Human Development and other partners. The collaboration is not a formal endorsement, and the layer is our own composite, with its inputs and weights in §5. It is a population-level index and must not be read as a clinical or individual diagnosis. It is not named after the research anywhere in the interface, because the name described what Legaz et al. measured rather than what we compute.
It describes environmental exposure, not residents. The layer measures the pressure of a place's surroundings; it makes no claim about the brains, capacities or worth of the people who live there. It must never be read as "this neighbourhood's residents have older brains." We are alert to the risk that a "brain ageing" frame laid over already-marginalised neighbourhoods can stigmatise them, so the layer is framed as exposure, makes no neighbourhood-level individual claims, and is suppressed where the underlying environmental data is weak.
Reproducibility differs by layer, and we are precise about it. The country layer is one-click verifiable by anyone, via the verify chips back to the published series behind every value. The within-city surfaces trace to the primary source named on each city's own card, and the data files the site serves carry the same values shown on the map, so a reader can check the map against the file rather than take it on trust. The derived layers (the environmental burden index, GHSL corrections and ground overrides) are editorial choices rather than measurements, which is exactly why this page documents what goes into each one and why (§5, §6). The claim-check engine ships with its own methodology, and rejects any computed claim whose numbers do not trace to a sourced value.
Each entry carries an independently written description rather than copied text, its category, its organisation, its place and a link out. The entries are drawn from three registers: the atlas's own curation, a feed built from the Pathfinders SDG16+ Inequality Solutions Portal, and a further feed built from Atlantic Fellows programme project pages. Every link is checked live before an entry is added. A project is placed by the city its work is based in, an organisation's own address or, for a national programme, its capital, and never at a personal address.
World Bank Data360 / WDI (with federated IMF and UNICEF series), World Inequality Database (WID.world), LIS Key Figures (open tier, cited as required; LIS microdata aggregates used in the claim-check prototype are held under a non-commercial data-use agreement), V-Dem, UNDP Human Development Reports, Our World in Data, Climate TRACE (CC BY 4.0), national census & statistics offices, IQAir, and ESA / NASA satellite programmes (Sentinel, VIIRS, FIRMS).
The scientific studies and research datasets this platform draws on, beyond its operational data feeds: