Running a field survey: the checks that save the data

A failed survey only shows at analysis, when it is too late to go back to the field. Here are the checkpoints that make the difference, in the chronological order of a collection.

Before the field

• Pilot the questionnaire on twenty real people. The pilot always reveals two or three questions nobody understands as intended.
• Translate, then back-translate. The question is translated into Hausa or Zarma, then translated back into French by someone else. The gaps between the original and the back-translation are the questions to rewrite.
• Set replacement rules. What does the enumerator do if the sampled household is absent? The rule must be written before departure, or each agent invents their own and the sample loses its validity.
• Plan for refusals. A refusal rate is not a failure: it is data. It must be recorded, not worked around.

During collection

• Check the first three days, every evening. That is when systematic errors take hold.
• Watch interview duration. A forty-minute questionnaire completed in twelve was not administered.
• Check GPS scatter. Ten interviews at the same point suggest forms filled under a tree rather than in the compounds.
• Call back five per cent of respondents. Two questions by phone are enough to confirm the visit took place.

At closure

• Clean before analysing. Impossible values, duplicates, inconsistencies between linked questions: the list of corrections must be documented, not applied silently.
• Keep the raw dataset. A cleaned dataset with no preserved original makes later verification impossible.
• Anonymise before release. Removing the name is not enough: village, age and occupation together often re-identify a person.

The budget line everyone forgets

Three items are missing from most survey budgets: the questionnaire pilot, quality control call-backs, and cleaning time. Together they are about fifteen per cent of the total cost — and they decide whether the data will be usable.