How should outliers be treated in normative analysis?

How should outliers be treated in normative analysis?

• A. Remove all outliers unconditionally.
• B. Always retain outliers to preserve data variety.
• C. Normalize all data to reduce outlier impact.
• D. Investigate potential data entry errors or protocol deviations; decide on retention or exclusion based on predefined rules.

Answer: (D)

Outliers should be handled by first examining their source and applying predefined rules. In normative analysis, you look for data quality issues—data entry errors, instrument glitches, or protocol deviations—that could explain extreme values. If such problems are found, you correct or exclude the affected data according to documented criteria. If the outlier is a valid observation under the measurement conditions, it should be retained, often with robustness checks to see how conclusions hold when it is included or excluded. Document the decision and use predefined rules to guide retention or exclusion, ensuring consistency and transparency. This approach is better than removing all outliers without basis (which can discard meaningful variation), or keeping every outlier (which can distort results), or simply normalizing data (which can mask underlying data quality issues and change interpretation).

Outliers should be handled by first examining their source and applying predefined rules. In normative analysis, you look for data quality issues—data entry errors, instrument glitches, or protocol deviations—that could explain extreme values. If such problems are found, you correct or exclude the affected data according to documented criteria. If the outlier is a valid observation under the measurement conditions, it should be retained, often with robustness checks to see how conclusions hold when it is included or excluded. Document the decision and use predefined rules to guide retention or exclusion, ensuring consistency and transparency.

This approach is better than removing all outliers without basis (which can discard meaningful variation), or keeping every outlier (which can distort results), or simply normalizing data (which can mask underlying data quality issues and change interpretation).