What Are Leverage and Influence?
You'll learn how individual data points can disproportionately affect your regression line, and why this matters for trustworthy analysis.
The Core Idea
In regression, not all observations are created equal. Some points have more power to pull your fitted line toward themselves than others.
Leverage refers to an observation's position in the predictor space—specifically, how far its X-value is from the mean of all X-values. High-leverage points sit far from the center of your data (think: extreme X-values). They have the potential to affect the regression line strongly, simply because of where they sit.
Influence is about actual impact—how much the regression line would change if you removed that observation. A point is influential when removing it substantially changes your slope, intercept, or predictions. High leverage doesn't guarantee high influence; a point far out on X can still align perfectly with the trend set by other points.
An Analogy
Think of a seesaw. Leverage is how far from the center you sit. Influence is whether you're actually making the seesaw tilt. Someone sitting far from the center (high leverage) could tilt it dramatically, but if they're balanced by others, the tilt is minimal. Someone close to center (low leverage) typically can't tilt it much, no matter their weight.
Why This Matters
A single influential observation can dominate your entire regression result—changing conclusions, predictions, and coefficient estimates. Identifying these points helps you decide whether they're genuine data or problematic outliers that warrant investigation.
Key Takeaway: Leverage describes where a point sits in X-space; influence describes how much that point actually changes your regression results when included.