# Scientist Excludes Outlier Data Point, Risking Research Integrity Questions
Tom Gauld's latest cartoon for New Scientist captures a recurring tension in scientific research: the temptation to remove inconvenient data points that don't fit expected patterns.
The cartoon depicts a researcher examining study results, identifying one measurement that deviates sharply from the rest. The researcher then simply removes it from the analysis, allowing the remaining data to produce cleaner, more statistically significant findings. This scenario plays out regularly in laboratories and academic institutions worldwide, raising persistent questions about research practices, statistical manipulation, and the boundary between legitimate data cleaning and problematic cherry-picking.
Outliers serve different purposes depending on context. Sometimes they reflect genuine measurement errors, equipment failures, or procedural mistakes that justify removal. A thermometer malfunction during an experiment, a contaminated sample, or a participant who didn't follow protocol all generate outliers for valid reasons. Removing these improves data quality and accuracy.
But other outliers carry real information. They may indicate unexpected phenomena, rare events, or effects that operate at study extremes. They can suggest limitations in current theories or hint at new biological mechanisms. By excluding them, researchers risk missing genuinely novel findings.
The problem intensifies when researchers selectively remove outliers only when they threaten desired conclusions. This practice, sometimes called "trimming" or "winsorization" without preregistration, amounts to p-hacking. It artificially inflates statistical significance, makes weak effects appear robust, and produces findings that fail to replicate in subsequent studies.
Professional guidelines address this tension. The American Psychological Association, the Committee on Publication Ethics, and most major journals recommend that researchers pre-specify outlier removal criteria before analyzing data. These criteria should depend on the measurement method and biological logic, not on whether results align with hypotheses. Researchers must report when and why data points were excluded, allowing readers to evaluate whether decisions were defensible.
The issue gained prominence after replication crises across psychology and medicine revealed how widespread problematic practices had become. Studies showed that vague data cleaning procedures contributed significantly to irreproducible results. Journals now increasingly demand transparency about data exclusion.
Gauld's cartoon works because it exaggerates something many scientists recognize: the human temptation to nudge results in favorable directions. It's not typically malicious. Researchers convince themselves that particular outliers "don't count" for technically defensible reasons, even when unconscious bias shapes those justifications. The cartoon serves as gentle satire while highlighting a real methodological problem.
Open science initiatives now push researchers toward registration, pre-analysis plans, and open data. These practices create accountability by forcing researchers to commit to analytical decisions before seeing results. They reduce opportunities for unconscious bias to influence which data survive final analysis.
Gauld's contribution to New Scientist reflects the publication's role in making science accessible and critiquing research culture. By highlighting methodological humor alongside serious science reporting, the magazine reminds readers that scientific integrity requires constant vigilance against our own cognitive blind spots.
The cartoon ultimately endorses a simple principle: let the data speak honestly, even when it contradicts expectations. That's harder than it sounds.