Hi all — I keep running into the same challenge and wanted to hear how this community handles it: knowing which submissions you can actually trust after the data comes in.
A few things quietly undermine a dataset — interviews done too fast to be real, duplicates, GPS that doesn’t match where the interview should have happened, or the occasional fabricated (“curbstoned”) response. At small scale you can eyeball it; at thousands of submissions it’s much harder.
I’d love to compare notes on the methodology:
- Which signals do you actually trust to flag a suspect interview — GPS, interview duration/timing, audio, response patterns, back-checks / re-interviews?
- How do you set thresholds without drowning in false positives?
- Where does it break down for you — enumerator training, supervision, or only catching it at the analysis stage?
- Does anyone build these checks into their Kobo workflow (validation rules, exports to another tool), or is it mostly manual review afterwards?
Not promoting anything — genuinely trying to learn what works (and what doesn’t) from people who live in Kobo every day. What’s your approach?
