How should outliers be handled in evaluation data?

• A. Always delete all data points that appear as outliers.
• B. Immediately apply a log transformation to all data.
• C. Investigate cause, consider transformation or robust statistics, or conduct sensitivity analyses with/without outliers.
• D. Ignore outliers and proceed with analysis.

Answer: (C)

Handling outliers involves understanding why they appear and how they affect your conclusions. Start by checking data quality: could a value be a data-entry error, a measurement issue, or is it a true extreme observation? If an outlier is a mistake, correct it if possible or justify removing it. If it reflects real variation, you shouldn’t automatically discard it.

Next, consider ways to address the issue without simply dropping data. A transformation, like a log or other scale, can sometimes stabilize variance and make patterns easier to detect, but it isn’t a one-size-fits-all fix and can change interpretation, so use it thoughtfully. Robust statistics offer another route: methods designed to be less sensitive to extreme values (such as robust regressions or using medians/trimmed means) let you reduce the influence of outliers without deleting them.

Finally, conduct sensitivity analyses—repeat the analysis with and without the outliers to see whether the main conclusions hold. If results shift substantially, you need to report that limitation and discuss how it affects interpretation.

This combination—investigating the cause, considering appropriate transformations or robust methods, and performing sensitivity analyses—provides a balanced, transparent approach to outliers. That’s why this option is the best choice.