When to use this
Use this before analyses that are sensitive to extreme values or influential cases.
Why it matters
Potential outliers can be data errors, valid rare cases, or influential observations; those possibilities require different responses.
Step-by-step instructions
1. Confirm coding and valid ranges.
2. Review distributions and flagged values.
3. Check the source questionnaire or collection logic when possible.
4. Document whether a value is erroneous, plausible, or uncertain.
5. Preview any exclusion or winsorisation.
6. Approve a new working version only with justification.
Expected outputs
Reviewable diagnostics and an auditable decision trail.
How to interpret the results
A flag is evidence for inspection, not an instruction to remove a respondent.
Common mistakes
Do not delete cases solely to improve significance or model fit. Do not alter the immutable source dataset.
Was this article helpful?
Your response helps the methodology team improve this documentation.