Sampling decisions made in an office determine how much of a survey survives contact with the field. A few design choices protect data quality when access is uneven.
Survey design is often treated as a questionnaire problem. In practice, the decisions that determine whether the resulting data can support conclusions are made earlier: how the population is defined, how the sample is drawn, and what will happen when parts of that sample cannot be reached.
Plan for non-response before it happens
Access problems are predictable in many settings — seasonal migration, weather, security, school calendars. A design that treats these as exceptions produces a dataset with unexplained gaps. A design that anticipates them specifies replacement rules in advance and records the reason for every non-response, so the analysis can describe who is missing.
Pilot the instrument where it will be used
Piloting in the capital rarely reveals the comprehension problems that appear in the actual survey districts. A pilot of even 25 to 30 interviews in representative locations typically surfaces translation issues, ambiguous response options and questions that take far longer than estimated.
Build quality checks into the schedule
- Same-day review of submitted forms during the first week of fieldwork.
- Independent back-checks on a defined share of interviews.
- Automated consistency rules applied at the point of data entry rather than at the analysis stage.
None of these steps is expensive relative to the cost of fieldwork itself, and each one reduces the chance of discovering an unfixable problem after the field team has demobilised.