Selection Bias

The distortion that occurs when the process of selecting a sample systematically excludes certain types of observations—producing a sample that doesn't represent the population. Survivorship bias (studying only successes), self-selection bias (people who volunteer for studies differ from those who don't), attrition bias (people who drop out of studies differ from those who stay), and Berkson's bias (hospital samples don't represent the general population) are all forms of selection bias. The fundamental error: drawing conclusions about the whole from a part that was selected in a non-random way.

When to use it

When conclusions drawn from available data might be distorted by how the data was collected; when success analysis needs to include failures to avoid survivorship bias; when research findings don't replicate (selection bias in the original sample); when any decision depends on data that was generated through a non-random process.

How it can help

For any analysis: map the selection process that generated your data and identify what's systematically excluded. Customer feedback: who responds to surveys? (The extremes—very happy and very unhappy.) Competitor analysis: which competitors are you studying? (The visible ones—missing quiet threats.) Performance data: which projects are in your dataset? (The completed ones—missing the abandoned.) The correction: either adjust your conclusions to account for the selection or supplement with data from the excluded population. Ask: 'what's NOT in this dataset, and how would including it change my conclusions?'