Survey Data Quality Checklist for Field Teams
Survey data quality depends on three phases: validation built into the form before launch, monitoring during collection, and verification before analysis. Teams that leave quality until post-fieldwork cleaning lose data they cannot recover. Most quality signals must be captured during collection.
Use the ODK Field Data Quality Toolkit when you need the same process in a shorter field-team checklist.
This is the checklist we use with ODK field teams. It is short on theory and long on things you can switch on this week.
Before fieldwork: build quality into the form
- Constraints on every numeric question. An age of 250 should be impossible to submit, not flagged later. Copy patterns from XLSForm constraint examples.
- Skip logic instead of enumerator judgment. Irrelevant questions hidden by relevant expressions cannot be answered wrongly.
- Required marks on critical fields. Consent, IDs, and outcome variables should never come back empty.
- Audit log enabled. Add an
auditrow with GPS and change tracking — it cannot be added retroactively to existing submissions. See the ODK audit log explained. - Pilot with real enumerators. Five practice interviews surface more form bugs than any desk review.
| type | name | parameters |
|---|---|---|
| audit | audit | track-changes=true identify-user=true location-priority=balanced |
The audit log is the foundation of every check in the next two phases. If you enable only one thing from this checklist, enable it — before the first real interview.
During fieldwork: monitor daily, not weekly
- Interview duration. A 45-minute questionnaire completed in 6 minutes is your loudest fraud signal. Watch the distribution per enumerator.
- GPS plausibility. Submissions should cluster where the sample lives, not at the field office or one enumerator's home.
- Submissions per enumerator per day. Sudden spikes near deadline are the classic curbstoning pattern.
- Answer-change patterns. Heavy late editing of key variables, visible in the audit log, deserves a phone call.
- Same-day review cadence. Catching a confused enumerator on day 2 saves the dataset; catching them on day 20 does not.
Doing this manually means scripting against raw audit CSVs every evening. This is the exact problem DataSnap automates: it ingests submissions live from your ODK server and turns duration, GPS, and audit signals into per-enumerator risk scores on a map.
Check my data quality freeAfter fieldwork: verify before you analyze
- Run a curbstoning screen. Cross-reference duration, GPS, and timing outliers; investigate the worst 5% of interviews first.
- Check enumerator-level distributions. One enumerator whose answers never vary is statistically loud — see enumerator performance monitoring.
- Reconcile counts. Submissions on the server versus the sample frame versus enumerator logs — gaps mean lost or fabricated interviews.
- Document exclusions. Every dropped interview gets a recorded reason; silent deletions are how datasets stop being defensible.
- Archive the audit logs with the dataset. Reviewers and donors increasingly ask for them.
The bottom line
Data quality is not a cleaning step; it is a collection design. Constraints and the audit log cost you an hour before launch, daily monitoring costs ten minutes with the right dashboard — and together they are the difference between a dataset you defend and a dataset you apologize for.
Check my data quality freeNext: How to detect curbstoning →
Frequently Asked Questions
How do you ensure data quality in surveys?
Build validation into the form before launch with constraints, skip logic, and an enabled audit log; monitor interview duration, GPS, and per-enumerator submission patterns daily during fieldwork; then screen for curbstoning and outliers before analysis. Quality signals must be captured at collection time.
What is the most important survey data quality check?
Enabling the ODK audit log before fieldwork starts. It records timing, GPS, and answer changes for every interview and cannot be added retroactively. Nearly every reliable fraud and quality check — duration analysis, location verification, edit patterns — depends on it.
How do you detect fake survey interviews?
Compare interview duration against the questionnaire's realistic minimum, verify GPS locations match the sample area, and watch for submission spikes from a single enumerator. Fabricated interviews are typically far too fast, wrongly located, and clustered in time.
What should I monitor daily during data collection?
Track interview duration distributions, GPS plausibility, and submissions per enumerator per day, and review them the same day. Daily review catches confused or fraudulent enumerators within days, while weekly review often discovers problems only after they have contaminated the dataset.
What is curbstoning in surveys?
Curbstoning is the fabrication of survey interviews by an enumerator who never visited the respondent. It is detected through audit log signals: implausibly short durations, GPS points away from the sample area, and unnatural submission timing patterns.
Can data quality checks be automated for ODK?
Yes. DataSnap connects to an ODK Central server, syncs submissions automatically, and computes duration, GPS, and audit-log risk signals per interview and per enumerator. It replaces manual nightly analysis of raw audit CSV files with a live dashboard.
No Credit Card Required
Build the form first. Add hosting when you need it.
Start free with LooprAI, then deploy a managed ODK Central server or run DataSnap checks when your project is ready.
Automation handles the mechanical work.
Researchers decide what the evidence means.