A translated questionnaire is a new instrument in each language. If a question means something different in Urdu than in English, the answers cannot be compared, however careful the fieldwork. A good translation workflow protects meaning, keeps the questionnaire's structure identical across languages and leaves a record of what changed.

1. Stabilize the source wording first

Translation works best from a source questionnaire that has already been reviewed and pretested. Translating every draft multiplies work in every language and lets inconsistencies creep in. Pretesting in the other languages may still reveal problems in the source. That is useful, but treat it as a controlled iteration: agree the change, update the source, send the change through the same translation review and record it.

2. Decide which languages you need and how they will be used

  • List the languages respondents actually speak in each area of the sample, not only the official languages.
  • Agree whether interviewers read a written translation or translate orally from another language. Oral translation on the spot varies between interviewers and is hard to check; a written translation is safer wherever it is feasible.
  • Record the language of each interview as a variable, so that language effects can be examined in analysis.

3. Translate and review as a team

A well-established approach in cross-cultural survey research is team translation, often described as TRAPD: Translation by two or more translators working independently, Review by a group that compares the versions, Adjudication of open choices by one responsible person, Pretesting, and Documentation of the decisions. It is not the only valid approach, and smaller projects may combine roles, but independent drafting, a review step and written documentation are worth keeping in some form.

Back-translation on its own is a weak check: a literal back-translation can look correct while the forward translation reads unnaturally or carries a different meaning.

4. Keep structure and codes identical

  • Translate labels, not codes. Choice codes, variable names and logic must be the same in every language, so an answer means the same thing in the data.
  • Keep the order of choices the same unless there is a methodological reason to change it, and document any exception.
  • Keep a glossary of key terms, such as household, head of household or income, and use it across the questionnaire and the training materials.
  • Translate interviewer instructions, hints, consent text and on-screen messages too, not only the questions.

5. Take care with inserted text and numbers

  • When a question inserts an earlier answer, such as a name, translate the whole sentence around it. Word order and grammatical agreement differ between languages, so the inserted value may need to sit somewhere else.
  • Check units, currencies, dates and number formats for each language and region.

6. Review after every source edit

Treat any change to source wording as a reason to review the matching translations: a translation that was right for the old wording may be wrong for the new one. Keep a version history, so you can tell which questionnaire version and language each interview used.

7. Test every language before fieldwork

  1. Ask a fluent speaker who did not translate it to read the whole questionnaire in each language, on the devices interviewers will use.
  2. Run complete practice interviews in each language, including routes through skip logic and inserted answers.
  3. Pretest with respondents who speak each language and ask how they understood the key questions (cognitive interviewing).
  4. Train interviewers in the language they will interview in, with the final translated questionnaire.

Further reading: the Cross-Cultural Survey Guidelines (University of Michigan), chapter on translation.