- Published
- Updated
An unusual value can be a data error, a genuine clinical finding or a difference in measurement conditions. Excluding an observation merely because it strongly influences the result can undermine scientific credibility. Review source evidence and context before deciding what it represents.
Define the question and assumptions
Distinguish impossible, unexpected and influential observations. Plan checks of units, dates, devices and source records. Prespecify rules for retaining valid observations and assessing their influence. Blinded review can help keep decisions independent of preferred comparative findings, but should not remove clinical context necessary to understand the measurement or evaluate its plausibility.
Hypothetical study example
In a hypothetical trial, an unusually high laboratory result affects the main estimate. Source review confirms it is correct. The team does not erase it to obtain a more balanced-looking result; it applies the planned method and appropriate supportive assessment without relabelling a genuine observation as a data error.
Design review checks
- Record source verification separately from assessment of statistical influence.
- Check unit conversions and measurement conditions at the individual observation level.
- Apply exclusion rules consistently across both treatment groups.
Record the decision and its limits
An accepted decision retains rationale and evidence for each affected record. Valid influential data are not removed solely for an inconvenient result. Additional analyses do not silently replace the main analysis; report their scope and sensitivity. Disclose timing and interpretive consequences if an outlier threshold changes after comparative results become available.
Frequently asked questions
- Is an outlier automatically an error?
- No. It may be a genuine finding or reflect its measurement context.
- Should an influential observation be removed?
- Influence alone is insufficient; scientific and prespecified rules are needed.