How the numbers get here, and what we do with them

Every place a country is given on this site is one of three things, and the word says which. Keeping them apart is the single most important habit on the site, because they are made by different people and mean different things.

Rank

A number a publisher printed. Shown exactly as published, with the publisher named. We never alter one and never call anything else one.

Position · derived

One published statistic, put in order by this site because its publisher ranks nobody. Always labelled derived, and never shown on a country profile, which promises that every rank there is the publisher’s.

Standing

A place in the Country Index or in one of its topics, worked out from several statistics together, and always named with its edition. Relative, and never a score.

The site publishes no score of its own, and it does not describe its results as official, absolute or endorsed. The Glossary defines the rest of the vocabulary.

When a statistic is imported, the value is parsed from the publisher’s file and written down exactly as it was published, digit for digit. Nothing is rounded, no float is tidied, no estimate is filled in for a missing year. This is a rule rather than a preference because rounding at import is permanent and silent: when the Human Capital Index was once stored to two decimal places, 172 economies collapsed onto 48 distinct values, and countries the World Bank separates were tied on this site. Ties are not cosmetic. They decide positions.

Precision is trimmed only when a number is drawn on screen, where the display chooses how many decimals suit the magnitude: a rank stays a whole number, a death count keeps its thousands separators, an index to four decimals is shown to as many as the column needs. That trimming is reversible. Rounding the data is not.

The one exception is a count of people. Deaths, displacements and populations are stored as whole numbers, because a fractional person is an artefact of a source’s storage and an integer is the honest type. Each importer that does this says so.

The rule above stops us rounding. It does nothing about a source that hands over numbers already rounded, and that has happened. Large platforms redistribute other agencies’ statistics, and some round them on the way: UNICEF’s under-five mortality estimates, published to fourteen decimal places with uncertainty bounds, arrive through the World Bank’s general data API at one decimal place. The World Bank’s own poverty headcount is published on its poverty platform to two decimal places and through the same API to one, so that Belgium’s 0.02 per cent arrives as zero.

So before any statistic is added, and again whenever it is re-imported, four things are done. The producer of the number is identified, as distinct from whoever is passing it on. One edition is pulled down both routes and compared row for row. If the producer serves more precision, the statistic is read from the producer, with its own importer. And the result is written down on the dataset’s page so nobody repeats the comparison. Every dataset on the site names where it was read from, and the Sources & publishers page lists the producers.

Where a publisher ranks the countries in its own report, that rank is shown and nothing else is derived from it. Where a derivation has been checked against thousands of the publisher’s own printed ranks and reproduces them exactly, the derived rank is stored for the years the publisher did not rank, and the dataset page says so. Otherwise no rank is stored, and the page says what the data lacks.

A statistic whose publisher ranks nobody may be given a position · derived by ordering its published value, but only when it deserves one: it must have a clear better end, and it must not pile up at a ceiling. Electricity access is the standard example of a statistic that gets none, because more than a hundred countries tie at 100 per cent and a position reading “1, 1, 1 … 123” is noise. The site’s rule of thumb is no tie deeper than about three per cent of the field. A statistic with no better end at all, such as manufacturing’s share of the economy, is never ordered and never coloured good or bad.

A standing is different in kind. It combines several statistics, which means it needs a way of putting values with different units on one footing, and that is the subject of the next section.

The Country Index is built in four steps, and the full working, with every number of the current edition, is on the index page. In outline:

1. A few key statistics vote. Each of the 12 topics has a small set of keys — 41 in the 2026 edition — chosen one at a time for the distinct concept each carries. Every other statistic on the site is context: charted, linked, downloadable, and without a vote. When everything voted, importing five more labour series quietly made the index more about labour.
2. Each value becomes a share of its own field. A country’s value on a key is turned into how far down that key’s list it sits, as a fraction from 0 to 1, with tied countries sharing the middle of the places they span and keys where a high number is worse turned round first. Dividing by the field first is the step that must not be skipped: 22nd of 44 and 108th of 215 are the same showing, and averaging the raw numbers would say loudly that they are not. Fields here run from about 44 countries to about 215, and averaging raw ranks would hand a quiet bonus to any country measured mostly in small ones.
3. Topics are averaged, not statistics. A topic standing is the plain average of a country’s key shares on that topic, with no weighting, because any weighting would be our opinion dressed as a measurement. The index standing is then the plain average of the topic standings: one topic, one vote.
4. Floors decide who is placed. A country needs readings on half of a topic’s keys, rounded up and never fewer than 2, for a topic standing; and standings on 75 per cent of the voting topics — 9 of 12 this edition — for a place in the index. Countries below a floor are listed unplaced with the count that excluded them. An edition is publishable at all only with 7 or more voting topics.

Region standings follow the same idea one level up: a region’s place among its peers is the median of its members’ positions, printed with its field, and never called a rank.

Naming a statistic as a key is an opt-in, not an admission. Each edition re-runs seven tests, and a key that fails one is demoted to context for that year, with the reason printed on the edition page. No rule is bent to fill a thin topic: a topic that cannot reach two admitted keys is published as a dashboard with the reason at the top.

1. It has a better end: a high value is clearly good or clearly bad.
2. It is a rate, share, index or per-person figure, not a raw count (which ranks countries by size), a growth rate (which ranks them by where they started) or a base-year index (which compares a country only with itself).
3. It measures at least 120 countries.
4. It describes a year inside the edition’s window, which reaches back at most 2 years.
5. No more than 10 per cent of its field ties at either end.
6. It does not repeat another key in the same topic. A rank correlation of 0.92 or more counts as repeating, as does being a component or an adjusted variant of another key. Judged within a topic, never across the whole index, because almost everything correlates with income.
7. Its licence permits derivative works, which is what a ranking is.

The average that orders the Country Index exists — it is in the edition file so the ranking can be audited — and it is never printed. Every composite that prints a number explains what the number is measured against; this one would have nothing anchoring it, so a figure out of 100 would read like a measurement somebody made when it is a mean of positions that moves with the field.

What is published instead is the position and the field it is a position of; a range, showing where the country would sit if any one key were left out; its coverage; and, on expansion, every underlying reading in the site’s fixed vocabulary — “rank 136 of 181, Transparency International” where a publisher ranks, “position 100 of 188 · derived” where this site ordered the statistic itself.

A country is named at the end of a measure — “highest”, “lowest”, a record, a place on a podium — only when it holds that value alone. Where several share it, the page says how many and names none of them first: “122 countries share 100%” rather than a winner picked out of a tie by the order the rows happened to arrive in. Thirty-three of the 174 countries ranked in the Country Index 2026 share their place with at least one other.

The same rule holds on a chart, where the entity may be a passport or an airline rather than a country, and on every statistic page. It is why some pages decline to draw a podium at all.

A statistic is published only when its publisher’s own terms permit reuse, including commercial reuse, with credit. Datasets whose publishers restrict reuse, or publish no terms at all, stay off the site: silence is not permission, because copyright is automatic. The licence and the date it was last checked are printed on every dataset page, and thirteen indices were removed in August 2026 when this rule was adopted and their terms were read.

Some of the key statistics carry a ShareAlike condition, which obliges anything built on them to carry the same licence. The Country Index 2026 is therefore published under CC BY-SA 4.0. Some key inputs are ShareAlike, so this derived ranking carries the same licence. The Terms of use say what that means for you.

The Country Index is published once a year and named for the year of publication. Because publishers number their reports differently, a key is chosen by the year its numbers describe, not the year on its cover, and the edition prints the honest span of years its readings cover — 2023–2025 for 2026 — rather than one tidy year.

A published edition is never changed. When the next one appears, the last is kept exactly as it was, so a citation keeps pointing at the numbers it was written about. Every edition is built by the same rules, set out in full on this page, so one edition can be read against another.

A method that lists no weaknesses has not been examined. These are the ones that apply to every edition; the ones particular to the current edition are listed on its page.

• It is relative. A standing says where a country places among the others measured alongside it. It can move because they moved, or because a publisher measured more of them.
• The publishers did not agree to be read together. Each key measures one thing well. Putting 41 of them side by side is this site’s decision, and the result is a construction, not a finding any of them made.
• A field is not the world. Dividing by the field removes its size but not its composition. A statistic measured across 130 countries says nothing about the rest, and a standing built on keys with different fields rests on slightly different worlds.
• Some subjects are thinner than others. Where the only measures of a concept are licensed shut, count rather than rate, or cover too few countries, the topic that needs them stands on fewer keys. The edition page names each gap.
• The window ages. Keys describing the oldest year the window allows must be re-imported before the next edition or they drop out of it.