High-frequency checks are routine checks on incoming survey data, often run daily while fieldwork is under way. Their purpose is practical: to find problems while interviewers are still in the field and respondents can still be reached. Problems found after fieldwork can still be investigated, and some can be corrected, for example by checking records or contacting respondents again where consent allows. Earlier detection makes the response more timely and reliable: interviewers can be coached, households revisited while memories are fresh, and the same problem prevented in later interviews.
What to check
Useful checks answer questions a field manager would ask anyway:
| Question | Example check |
|---|---|
| Are we on track? | Completed interviews per day, area and interviewer against the plan |
| Is every interview unique and complete? | Duplicate household or respondent IDs; key sections left empty |
| Were procedures followed? | Consent recorded before substantive answers; eligibility rules met |
| Do answers make sense together? | Contradictions between related answers, such as a child older than the parent |
| Are some values unusual? | Values far outside what most interviews report |
| Do patterns differ between interviewers? | Shares of very short interviews or of “Don't know” answers, compared across interviewers |
Set thresholds you can defend
- Base thresholds on evidence such as pilot data, earlier survey rounds or published benchmarks, not on round numbers chosen in advance.
- Allow for the questionnaire's design: skip logic makes some interviews legitimately much shorter or longer than others.
- Write down every threshold with its reason, so that reviewers, and later readers of the survey report, can see why a case was flagged.
- Review thresholds after the first days of fieldwork. Adjust them openly and record the change.
A flag is a signal for review
A flag says that a case deserves a closer look, not that it is wrong. An unusually short interview may come from a household with few eligible members; an extreme value may be true. Every flag should lead to a human step: checking the record, asking the interviewer or supervisor, calling the respondent back where consent allows, and recording the outcome. When a value is corrected, the correction should be traceable and the original value kept.
Organize the follow-up
- Share results quickly and in a form field teams can act on: by interviewer, area and day.
- Give each flag an owner and track it until it is resolved.
- Discuss interviewer-level patterns privately and constructively. They are a basis for coaching, not an accusation.
An example: interview duration
The figures in this example are illustrative, not recommendations. Suppose that in a pilot most complete interviews take 35 to 60 minutes, and households without children, who skip a long module, take about 25. The team flags complete interviews under 20 minutes or over 120 minutes for review and looks separately at each interviewer's share of short interviews. In the first week, one interviewer has many short interviews. Reviewing them shows that the child module is being ended early; after coaching, the interviewer's durations return to the expected range.
Limitations
- Checks find unusual data, not all wrong data. A plausible but false answer passes every check, which is why back-checks and supervision remain necessary.
- Comparisons between interviewers need enough interviews per interviewer to mean anything, and interviewers working in different areas may legitimately differ.
- Every check creates review work. A few well-justified checks that are followed up are worth more than many that are ignored.
Further reading: J-PAL's research resource “Data quality checks” and the World Bank DIME Wiki page “Monitoring Data Quality”.