The Best Time to Catch a Data Problem Is Day One

One thing I've learned over the years is the importance of checking your data early, from the first day of fieldwork. Often, you’ll find something unexpected, and catching something early is always better than catching it late.

On a recent agricultural survey in Uganda, I ran a paradata check on the first day of data collection. (Paradata is the data about your data: timestamps, survey durations, interviewer IDs). Most people don't look at it until something goes wrong. I've learned to look at it from the start.

Three things stood out immediately. Several interviewers had average survey durations of over four hours, while others were completing interviews in under twenty minutes. And some interviewers appeared to have started multiple separate surveys within minutes of each other.

None of these patterns are automatically a problem. But all of them are worth understanding.

So, I flagged them to the fieldwork manager and asked for explanations. Working through each one with the field team, the reasons became clear: soe interviewers were leaving forms open when a respondent wasn't ready and returning later; in other cases farmers had been gathered together by local mobilisers, and interviewers were beginning surveys with all gathered farmers at once; and software issues were causing timestamp anomalies during interrupted surveys. In each case, the context made sense once we understood what was happening on the ground.

These conversations matter beyond just data quality. They help build a working relationship with the field team based on transparency rather than suspicion, and signal that data quality is taken seriously at every level. In this case no data was discarded, no interviewers were pulled. But we now understood our dataset better, and the field teams knew that someone was paying attention.

I've published research on this topic, specifically on how paradata can be used to monitor and improve data quality during fieldwork. The academic argument and the practical reality are the same: the best time to identify a problem is before it has had time to replicate itself across your entire dataset.

It doesn't take long. But building such checks into your fieldwork routine from day one can have huge payoffs.