Bortkasting

Bortkasting is a term used in statistics and data analysis to describe the practice of discarding data points from a data set before analysis. The aim is to remove erroneous measurements, reduce the impact of extreme values, or otherwise improve the quality of the data for modelling and inference. In English-language terminology, the concept is often referred to as data cleaning or outlier removal, though bortkasting can cover a range of practices from simple exclusion to more structured procedures.

Common approaches include excluding observations outside a plausible range, removing a fixed percentage of observations from the tails (trimming), or replacing extreme values with less extreme, but still plausible, values (winsorizing). The criteria for bortkasting should be determined before analysis and documented to avoid introducing bias.

Bortkasting is controversial. While it can reduce noise and improve the performance of statistical methods, it can also bias estimates, particularly if extreme values carry information about rare events or the underlying data-generating process. In some disciplines, robust statistical methods that accommodate outliers without discarding data are preferred.

In practice, bortkasting should be transparent, justified by data collection errors or methodological considerations, and clearly reported in study methods. It is distinct from censoring or truncation, where values are not observed due to the design of data collection rather than being removed by analyst choice. See also: outliers, data cleaning, trimming, winsorizing, robust statistics, preprocessing.