Insights
Data quality is won or lost in the field
Every organization I have worked with has a version of the same story. Months after a survey, someone opens the dataset to write the report and finds ages of 250, households interviewed twice and a district where every respondent apparently gave identical answers. By then the money is spent and the teams have gone home. Nothing can be fixed. It can only be footnoted.
The lesson took me one season as an enumerator to learn and I have watched it hold on every project since: data quality is not a cleaning task. It is a collection task. Whatever survives the field is what you will analyze, and no amount of clever cleaning creates information that was never captured.
Three checks that prevent most of it
First, constrain at the source. Your data collection form should refuse impossible answers before they exist. Age limited to a sensible range. Skip logic that hides questions that do not apply. Required fields that cannot be left blank. Ten minutes of form logic saves ten days of cleaning.
Second, review daily, not at endline. Someone should look at each day's submissions the same evening: counts per enumerator, duration of each interview, patterns that look copied. A three-minute interview for a forty-question survey is not speed. It is fabrication, and you can only act on it while the team is still in the field.
Third, close the loop with enumerators. When a check flags something, the enumerator should hear about it the next morning, kindly and specifically. Teams that know someone is reading their data produce better data. Teams that suspect nobody looks produce whatever gets them home earlier.
Why it matters
Decisions ride on this data. Budgets shift, programs scale or close, communities receive support or do not. Treating quality as a field discipline rather than an office cleanup is the single cheapest improvement most organizations can make to the evidence they run on.
If you want help setting up daily checks for your next survey, book a session and bring your form. We will pressure-test it together.