A statistical phenomenon where combining data reverses the apparent story. The article explores Simpson's paradox using a medical treatment comparison. Treatment A shows higher success rates than Treatment B for every patient type, yet Treatment B wins overall.
This occurs because Treatment A was given mostly to higher-risk patients, while Treatment B was given mostly to lower-risk patients. The overall result reflects both treatment outcomes and group composition. The paradox demonstrates how data visualization can be misleading without considering underlying variables.
The author emphasizes the importance of careful analysis and visualization to avoid drawing incorrect conclusions from aggregated data.
Source: Towards Data Science · Summarized by HeadlinesBriefing