The Survey Data Quality Playbook

Before you trust a single average or percentage, check whether the responses behind it are actually clean. A quick pass catches most problems.

1. Remove duplicate responses

Check for a duplicate response from the same respondent - most common with public links shared multiple times, or a respondent who submitted twice after an error message.

2. Flag speeders

A response submitted in a fraction of your median completion time was probably rushed through without reading the questions. Sort by time-to-complete and review anyone well below the pack.

3. Catch straight-liners

In any matrix question, watch for respondents who pick the same column for every row - it usually signals disengagement rather than a genuinely uniform opinion.

4. Handle missing data deliberately

Decide upfront whether a skipped optional question means "excluded from that question's average" or "counted as a non-response" - mixing the two silently changes your numbers depending on which questions people skipped.

5. Look for outliers, then investigate before removing

An outlier might be noise, or it might be your most important data point - a furious customer's real response shouldn't be dropped just because it's extreme.

6. Document what you excluded

Keep a short note of how many responses you removed and why, alongside your final numbers - "412 responses, 18 excluded as duplicates or under 10 seconds" is one sentence that makes your results defensible.