Course contentsShow
← Mathematics
Lesson 3087 of 3,12160. Statistics — ModelingFree lesson

Influential Points and Leverage

Identify high-leverage points that can disproportionately affect regression estimates based on predictor values.

Influential Points and Leverage

What You'll Learn

You'll learn how to identify observations that have outsized power to change your regression line due to their unusual predictor values.

The Core Idea

Imagine you're balancing a seesaw. A person sitting near the fulcrum has little effect on balance, but someone at the far end can tip the whole thing dramatically—that's leverage in regression.

In linear regression, not all data points influence the fitted line equally. Leverage measures how far a point's predictor value(s) lie from the center of the data. Points with high leverage are outliers in the predictor space (unusual X values), giving them the potential to "pull" the regression line toward themselves.

However, high leverage doesn't automatically mean high influence. A high-leverage point that falls near where the line would naturally be doesn't change much. But a high-leverage point with an unusual response value (unusual Y given its X) can drastically alter the slope and intercept—this is a truly influential point.

Analogy: Think of voting power. A swing voter in a close district (high leverage) only matters if they vote differently than expected (becomes influential). A voter in a landslide district (low leverage) has little impact either way.

Why This Matters

  • Leverage is determined purely by predictor values—it's calculated before looking at responses
  • Influence combines leverage with how far the point deviates from the pattern
  • High-leverage points deserve scrutiny: Are they valid data or errors? Do they represent a different population?
  • Removing or correcting a single influential point can dramatically change your regression coefficients and predictions

Always examine high-leverage points through residual plots and consider their impact on your model's conclusions.

Key Takeaway: High-leverage points have unusual predictor values that give them potential to disproportionately affect the regression line; they become influential when they also have unusual response values.