📐 Languages method

CLDR gives each country's share who speak a language. Speakers is that share times the country's population, summed over countries. Bilinguals count in every language they speak, so languages overlap and family totals overlap too.

A language's power is the Metro Power score of every metro where it is used, weighted by the share who can hold a conversation in it. A bilingual city counts in full for each language, so shares overlap and are never split.

Maps use a 1:110m world outline, which leaves out small territories. They are in the tables and not drawn. A click on a shaded country opens its page.

Reach is the share of residents who can hold a conversation in a language, first or second. The first-language share is native. The difference is vehicular: people who learned it to get by. The hub keeps native and reach rankings apart, since they answer different questions.

Ten lingua francas are tracked: English, Spanish, French, Arabic, Russian, Mandarin, Portuguese, Hindi and Urdu, Malay and Indonesian, and Swahili. A metro's reach is its state's reach, inherited and shown as inherited.

Tier 1: Census ability tables

A national census asked who can speak the language. The strongest figure, and the only one that can split first and second language.

Tier 2: Surveys

A representative survey, such as Eurobarometer. Good for comparison inside one survey, weaker across surveys.

Tier 3: CLDR country figure

Unicode CLDR gives the share of each territory that speaks a language, first or second, with its own cited source. Country grain, so every state of a country inherits it.

Tier 4: Modelled

A country figure adjusted by what a state's administrative language is. Always flagged, never ranked against a census figure on a board that hides the tier.

Today every reach figure is Tier 3. Census and survey tables replace the country figure for the states they cover, and the tier column changes with them.

CLDR shares overlap, so a bilingual country's shares add to more than 100%. The hub does not scale them down. Effective languages (1 divided by the sum of squared shares) and Greenberg diversity (1 minus that sum) use the raw shares.

A state's first language is its de facto label, else administrative, else national, from the States sheet. Dominance is its share where a census exists. Until then the board shows the country-level CLDR share and says so in Basis.

CLDR lists only each country's larger languages, so a diaspora language often has no figure: Arabic in the United States, France or Germany. Where CLDR is silent, a national census (Tier 1, summed from its state figures) or an official national survey (Tier 2, for countries whose census asks no language question) fills the gap, labelled with what it counts. A census or survey figure never replaces a CLDR one, and diversity and each country's top language use CLDR alone.

Where nothing current is measured at all, the hub makes its best estimate and says so (Tier 4, modelled). Arabic in France is the first: about 3.9 million speakers, 5.6%, within a range of 3.7% to 9.3%. One method counts Maghreb-born immigrants and their French-born children (INSEE) at the rates at which they speak the family language (INSEE and INED's Trajectoires et Origines survey, 2019 to 2020), with a stated Arabic-or-Berber split; the other scales INED's 1999 count of 3 to 4 million by the growth of the Maghreb-born population since. A measured figure replaces the estimate when one exists.