“We will check the data” is not a quality plan. A good plan makes quality observable. It identifies how the study could fail, what signals would reveal the problem, who reviews them, and what happens when a threshold is crossed.
Design controls around failure modes
Different methods fail differently. An intercept study may drift toward convenient respondents. A long CAPI survey may attract speeders or interviewer shortcuts. B2B recruitment may produce delegates instead of decision-makers. Name the specific risks before selecting controls.
- Sampling and quota drift
- Duplicate, coached, or ineligible participants
- Suspicious duration or response patterns
- Location and interviewer anomalies
- Translation or logic failures
- Incomplete evidence and unclear exception handling
Combine technical and human review
Metadata can flag unusual behavior, but context determines what it means. A short interview is not automatically invalid; a repeated pattern across one interviewer may be. Pair automated checks with audio review, back-checks, open-end reading, and field-manager observation.
Keep the decision trail
Every material exclusion or corrective action should be traceable. The final report should summarize what was checked, what was flagged, how it was resolved, and what remains as a limitation. That record is part of the deliverable, not internal housekeeping.
Strong research design includes the operational choices that determine who is reached, what is understood, and which evidence survives delivery.