Winsorization replaces extreme observations with less extreme boundary values. It offers an alternative to removing outliers or transforming the data when extreme values could affect a linear regression. section 4.2.1
Apply the technique as part of a documented outlier-treatment decision:
- Check whether observations exceed chosen boundaries, such as three standard deviations from the mean. section 4.2.1
- Replace qualifying values so they fall within a specified percentile range rather than deleting rows. section 4.2.1
- Compare the implications of winsorization, removal, and transformation, then record the reasoning for transparency and reproducibility. section 4.2.1
Screening the mds data
For stage1, the mean is 194.55 and standard deviation is 29.62; for stage2, they are 251.45 and 51.11. Python threshold checks found no observations beyond three standard deviations in either column, so no winsorization was required. section 4.2.1
