Languages

Interface text — buttons, menus, this paragraph — can be machine-translated. A clumsy rendering is an annoyance, nothing more.

The 45 poll statements are not interface text. A translated attitude question is a different instrument. If "should" becomes "must", or a word carries a political charge in one language it lacks in another, then the gap between two nations is partly translation and partly opinion — and no amount of statistics can separate them again. That gap is the entire product.

So poll items follow the model used by Pew, the ISSP and the World Values Survey — translate, review, adjudicate, pretest, document — and each translation carries the grade it has reached:

Roughly a third of the world's adults did not finish secondary school, and a very large share of respondents read DataRadio in a second or third language. A question pitched at university level does not measure opinion — it measures education, and reports it as though it were politics.

So every statement is written to a hard budget and the test suite fails the build if any item breaches it. The current English bank averages grade 4.5 and nothing exceeds grade 8. Translations inherit the same budget.

We deliberately write to the floor rather than shipping a different, simpler wording to poorer nations. Serving country A an easier question than country B would put a wording difference exactly where a national difference is supposed to be — which is the artefact this whole page exists to prevent.

Translating or reviewing a language you speak natively is the single most valuable contribution to this project. Write to [email protected] — reviewers are credited unless they'd rather not be.