August 19, 2026
Earlier this month, Stone Center Senior Scholar Branko Milanovic updated his All the Ginis (ATG) dataset. It now covers the period from 1950 to 2024 and includes 6,762 Ginis from 183 countries or territories. In this interview, Milanovic discusses the sources of the dataset, the changes in this recent update, and the continuing challenges in improving the production and collection of data on national income inequalities. He also gives a warning about technological change.
How would you describe All the Ginis in simple terms?
Milanovic: All the Ginis is a compilation of income and consumption Ginis collected, harmonized, and made available from various databases or platforms, or calculated by me from existing data sources, like the Luxembourg Income Study Database (LIS), the data center in Luxembourg. LIS plays a key role in ATG because it provides microdata from more than 1,000 nationally representative household surveys, so researchers can themselves decide which concept (income or consumption) and which form (gross or net; that is, before or after direct taxes) they want to use. They can also decide if the units over which inequality is calculated should be individuals or households. But ATG of course includes Gins from other large data sources, in particular from the rapidly expanding World Bank’s Poverty & Inequality Platform (PIP), which in principle covers all the countries in the world and provides, for most of them, income by each percentile of the distribution as well as Ginis. There are other platforms from which I compiled Ginis, including the World Income Inequality Database (WIID) at UNU-WIDER, and the Socio-Economic Database for Latin America and the Caribbean (SEDLAC), which is produced by the National University at La Plata in Argentina. Both of these platforms are excellent.
The key point to realize is that most of the Ginis included in ATG are calculated from household surveys’ microdata. As I mentioned, ATG very clearly shows if a given Gini pertains to income or consumption, if it is per capita or per household, and if income is net (of direct taxes) or not. So the data are, I believe, very well documented: What they measure and where they come from are clearly shown. ATG differs from some other databases that are either not sufficiently transparent (we do not know where their numbers come from) or rely extensively on imputations and extrapolations. ATG doesn’t do any of that, whereas, for example, the Standardized World Income Inequality Database (SWIID) does.
Finally, I should clarify that ATG deals only with inequality of income or consumption — that is, of annual flows. If someone is interested in inequality of stocks, which is wealth, they should consult probably the best and most extensive collection of wealth data in the world: the GC Wealth Project, which, as you know, is a project of the Stone Center and Roma Tre University. That data compilation has been created and managed by Salvatore Morelli and his many associates in New York and Rome.
How have things changed since 2004, when the original version of ATG was created?
Milanovic: When ATG was created more than twenty years ago (and there have been four updates since), there were not as plentiful data on income distribution as what we have now. Obviously, the World Inequality Database (WID), at the Paris School of Economics, has done a great job in making many more data available. Their data, however, are mostly fiscally-based, which means that for most countries in the world that lack significant systems of direct taxation or have introduced them only recently, WID does not have the data. Or they have the data for a year or two or three. That then leads to extrapolations and the use of various assumptions, and of course to the combinations with survey and administrative data, which may or may not be fully justified. Yet clearly WID has contributed a lot to our collective knowledge. WID’s preferred measure of inequality is also different: the share held by the top one percent or any other top group, rather than the Gini. This is an eternal debate. Which is better? It depends on your objective: If you want to focus only on the top, then surely the share of income received by top groups is very important. But note that, by definition, the top one percent tells you nothing about what happens to 99 percent of the distribution — that is, to practically everybody. Say that the top one percent gets relatively poorer (receives a lower share of total income), but that income inequality among the 99 percent become much greater. We know nothing about the latter part if we look only at the top. The Gini, however, by contrasting each person’s income against another person’s income (succession of bilateral income comparisons), is a measure that takes everybody into account. It runs, I should have said before, from 0, a theoretical situation where everyone has the same income, to 100, another theoretical situation where one person or one household has all the income. But it is true that Gini is a bit abstract: It reduces distributions to a single number, and that is tough for most people to interpret.
To go back to your question and conclude: Things have vastly improved in terms of our knowledge of income inequality since 2004. ATG has now many more competitors, and of course gives just one slice of the overall picture of inequality.
What would you like to improve in the next versions of ATG?
Milanovic: I would like to improve our knowledge about income distributions in the past. It took up to the 1980s to have about 20 countries included annually. There was an increase in coverage over the next two decades, and by the year 2000, there were about 80 countries included each year. That number reached 100 by the turn of the century. But having 20 or fewer data points annually in the 1960s or 1970s is really not enough. We know (I know!) that there are quite a lot of tabulated data (not microdata, because most old microdata have been lost) published on paper, in countries’ official income distribution publications or statistical yearbooks. To give you an example, ATG includes the data from different CEPAL (UN Economic Commission for Latin America and the Caribbean) reports from the 1950s and 1960s. I found these publications and added their Ginis to ATG. But there are many other such publications. If a person (a research assistant) interested in this type of work could be found who is interested in working in libraries to try to “excavate” such data, we could make substantial progress, especially with the data from the 1950s and 1960s. These were the years when both Western and Eastern European and Latin American statistical agencies began to field regular household surveys.
In the ATG version before the last I introduced so-called independent (INDIE) Ginis. Thay are generally Ginis that cover a number of consecutive years, come from the same or similar sources, and are calculated by the same researcher or organization. This gives them a consistency that compilations, however hard we try, do not always have. I would like to mention that, in this version of ATG, I include a very long series for the U.S., based on official Census Bureau Ginis that run annually from 1950 to 2024; an extraordinary series for Iran that goes from 1984 to 2024, thanks to the work of Djavad Salehi-Isfahani; and an equally excellent and consistent series of UK data from 1961 to 2014 provided by Jonathan Cribbs from the Institute for Fiscal Studies in London. There is also a great series for China: forty years, from 1984 to 2024. For China, moreover, ATG gives both consumption and income Ginis. Finally, there are long series for Brazil, Russia, and India.
And what would you like to improve that does not depend on you — that is, regarding issues of data production or collection?
Milanovic: It is an easy question. Ideally, I would like to improve the data from the 19th century. There is a project going on right now (you can see the description of the project in this working paper) run by Philipp Erfurth, Maria Gomez León, Giacomo Gabbuti, and me to collect and standardize social tables from the 19th century, and thus produce Gini estimates. Currently our knowledge of within-country inequalities in the 19th century is extraordinarily patchy. We have to rely there on social tables (listings of salient social groups with their estimated mean incomes), since there were neither fiscal data nor household surveys then. But such social tables have been done for relatively few countries (England is a country that began the practice with the social table done for 1688). Thanks to the work of extraordinary young researchers like Javier Rodriguez Weber and Carlos Castañeda we have such “dynamic” (that is, annual or almost annual) social tables for the 19th century’s Chile and Mexico; likewise, thanks to the work of Stefan Nikolić, Filip Novokmet and Piotr Paweł Larysz, we have data for Czechoslovakia and Bulgaria; and thanks to Peter Lindert and Jeffrey Williamson for the key 19th century dates for the United States. There are others, too; I can’t go over all the names of these excellent researchers. But all of this needs to be, first, put together. And second, many more of such tables need to be produced.
Another problem is that of the present. Recently, over the last decade or so, we have had a significant reduction in the number of household surveys conducted in many African countries. In some cases, this is due to conflict or war. In others, it is due to lack of administrative capacity and/or to a lack of interest by the World Bank to push countries harder to collect the data. It will be potentially a big problem. Africa is the only continent with a rising population, and we may know less about income of that population today than ten years ago! We lack good, or any, data for Nigeria, Congo, Sudan, and Tanzania. These four countries alone have 450 million people. This is like the entire European Union.
And there is also the perennial problem of countries like Saudi Arabia and North Korea that have never conducted income surveys. Or countries like Algeria that have, but refuse to share the data with anyone. Or Cuba, which has stopped producing such surveys since the 1990s. And unfortunately, countries that have been in civil or intranational wars, e.g. Libya, Syria, Afghanistan, Palestine. Ukraine obviously cannot realistically do surveys under wartime conditions. The truth is sad: Coverage of national income inequalities is not improving.
Any last thoughts?
Milanovic: An important note of caution for younger researchers. Digitalization or AI is not a panacea. In my lifetime I have seen dozens, if not hundreds, of surveys and datasets that have been lost. They disappeared because of incompatibility between different operating systems. Fast technological change means that obsolescence is very fast too. Unless you convert the data in time — and this is seldom done because it is expensive and people do not think about it — the data will be lost. It is simply a fact that the data on paper, including tabulations, are much more durable than the better but more fragile data on platforms that quickly become obsolete. So beware of, and prepare for, fast technological change!
Read More:
