National inequality averages can be accurate and still miss where inequality is lived. Housing, transport, employment, education and access to public space are organised through cities and regions, and the chart above already hints at why: metropolitan evidence brings that experience closer to view, but only where the underlying places are measured consistently enough to compare.
This article therefore makes two arguments at once. City-level data is essential because national stability can conceal local difference. The available international city data is also uneven enough that coverage must be treated as evidence, not as a footnote beneath the "real" result.
The visible world is shaped by who gets measured
The chart opening this piece counts city-year observations by country. Colombia and Kazakhstan dominate the available records, followed by a steep decline into countries with much thinner coverage. A country with more observations can appear more variable simply because more of its cities and years are visible.
This creates a policy asymmetry. Places with strong statistical systems generate the evidence needed to diagnose problems and evaluate interventions. Places with sparse data can disappear from comparative work regardless of how urgent their urban inequality may be.
Repeated measurement is concentrated in a small set of places
Figure 2. City-level monitoring depth is uneven, with a limited group of cities observed repeatedly and many represented only a few times.
Each city in the source dataset was counted by how many of the possible eight observation-years it actually has data for, then ranked, so "coverage" here means genuine repeated measurement of the same place rather than simply appearing in the dataset once.
The city-level coverage chart shows that national observation counts can also be misleading. Robust trend analysis requires the same place to be measured repeatedly. In this dataset, only 43 cities have the full eight years of observations, and 39 of those are in Colombia or Kazakhstan. Two countries therefore supply roughly 91% of the deepest longitudinal sample.
That concentration does not make their data less valuable. It makes broad global generalisation less defensible. A trend estimated from well-observed Colombian and Kazakh places cannot stand in for cities that appear once or not at all.
The latest observed levels span radically different realities
Figure 3. The lowest and highest latest available city observations reveal a wide inequality range across the 2010 to 2017 source window.
The most recent available Gini observation for each city was taken (not an average across years, since coverage is too uneven for that), then the extreme highest and lowest values across the whole set were pulled out and labelled directly.
The extremes give the coverage question substantive meaning. The lowest latest values in the dataset are concentrated in Kazakhstan, including Mangistau at 0.165. At the other end, Sana'a records 0.680, while Johannesburg and Tshwane reach 0.670 in their latest available observations.
These are not a contemporaneous global league table: the latest year differs by place, and the labels mix cities with regions and administrative units. They are evidence of scale. Urban inequality in the observed dataset ranges from comparatively compressed distributions to profoundly unequal ones, and national averages cannot convey that span.
Countries contain different urban distributions
Figure 4. Data-rich countries occupy different inequality bands and show different amounts of variation across their observed city-years.
Only countries with at least 20 city-year observations were kept, and their Gini values were rendered as boxplots side by side, so both the national median and the spread of cities around it could be compared on the same axis without a thin-data country distorting the picture.
The boxplots compare countries with at least 20 observations. Kazakhstan sits at the lower end of the observed Gini distribution; Colombia and Honduras sit much higher; the United States and Vietnam occupy intermediate positions with their own spreads. Within-country variation is visible as well as the separation between national medians.
This is the strongest argument for metropolitan evidence. A national statistic describes the centre of a country; city data describes the range of places around it. Housing markets, industrial structure, migration and public services can make two cities under the same national policy regime experience inequality very differently.
Time adds direction, but only where coverage permits it
Figure 5. Median Gini trajectories for the best-covered countries are shown with the range across their observed cities.
Only the best-covered countries were plotted as time series, with a shaded band around each median line showing the spread across that country's own cities, so the chart shows uncertainty explicitly rather than implying a false precision the underlying coverage can't support.
The time series introduces movement without pretending every line is equally reliable. Colombia remains high but trends modestly downward across the observed period, while Kazakhstan remains substantially lower. The shaded ranges show that cities inside the same country do not move as one unit.
The limited and uneven years are part of the interpretation. A smooth line over eight observations is not equivalent to continuous monitoring, and a two-year fragment should not be described as a long-run trend. The chart is strongest where it makes both the central tendency and the uncertainty created by city variation visible.
Inequality is also a measurement system
Figure 6. A large-format synthesis board that brings the article's five principal pieces of evidence into one readable argument.
The synthesis begins with an uncomfortable fact: the cities easiest to compare are the cities measured most often. Coverage, repeated observation, extreme values, within-country distributions and time trends all depend on that uneven evidence base. A clean global ranking can therefore express data availability as much as lived inequality.
The correct response is not to abandon comparison, but to preserve the uncertainty around it. Cross-sectional differences identify places worth investigating, while repeated local series are better suited to judging direction and policy change. Better urban decisions require both better outcomes and a more representative measurement system.
Better urban evidence changes what policy can see
Across the five views, coverage and outcome cannot be separated. Country counts show who is represented. City monitoring shows where trend analysis is possible. The extremes show the substantive range of inequality. Country distributions reveal local heterogeneity. The time series shows how patterns move where repeat observation exists.
City data is not a smaller version of national data. It answers different questions: where opportunity is concentrated, which metropolitan areas diverge from national norms and whether local policies change the trajectory. But those answers are only as credible as the consistency of place definitions and measurement.
The coherent argument is simple: national averages need a metropolitan lens, and the metropolitan lens needs a serious coverage audit. Better city policy begins by seeing both the inequality and the blind spots around it. The 91% figure is the one I'd want a reader to remember longest, not because it's the most dramatic number in the piece, but because it's the one that quietly limits how confidently any of the other numbers should be trusted.
Methods and original sources
The source contains 726 usable city-year Gini observations from 2010 to 2017, analysed in Python with pandas handling the coverage counts and country filtering and matplotlib rendering the bar, boxplot and time-series charts. Samples are highly uneven across countries and years, and the "City/region" field includes some oblasts, counties and other non-city units. The supplementary extremes and country-distribution figures are derived from the same source dataset. Results should not be treated as a complete global or national inequality ranking.