An outlier with long arms (extreme x) can drag your regression line across the room. That’s leverage… and it talks louder than the rest. Before you trust r or a slope, label the loud points and ask three questions: (1) Is it real? (2) Does it represent a scenario we care about? (3) What happens if we report both with and without it? You don’t have to exile unusual points to be honest; you have to show their influence and narrate the decision. The scatterplot is court; the outlier gets a hearing.
