Visualizing Leverage and Influence
You'll learn how to create visual diagnostics that help spot problematic observations based on their leverage and influence.
Why Visualize These Metrics?
You've learned to calculate leverage (hat values), Cook's Distance, DFFITS, and DFBETAS. But staring at tables of numbers isn't the best way to spot problems. Visualizations let you see patterns instantly and identify which observations deserve closer inspection.
Three Key Diagnostic Plots
Influence Plot (Bubble Plot)
This plot shows studentized residuals vs leverage, with bubble size representing Cook's Distance. Points in the upper-right or lower-right corners (high leverage + large residual) are trouble. Large bubbles anywhere signal high influence.
Analogy: It's like a radar screen where dangerous objects appear in specific zones and glow brighter (bigger bubbles) when they're more threatening.
Leverage-Residual Plot
A simpler version plotting leverage vs residuals without bubble sizing. Helps you quickly spot high-leverage points that also have large residuals—the combination that creates influence.
Cook's Distance Plot
Plot Cook's Distance for each observation (often as a bar chart with observation index). Values above common thresholds (like 4/n or 1) flag influential points immediately.
What to Look For
- High leverage + small residual: Not necessarily problematic (fits the model well)
- Low leverage + large residual: Outlier but limited influence
- High leverage + large residual: High influence—investigate these carefully!
- Large Cook's Distance: Point substantially changes the regression line
These plots answer: "Which observations am I most worried about?" at a glance.
Key Takeaway: Influence plots, bubble plots, and leverage-residual plots turn diagnostic statistics into visual patterns, making it easy to identify observations that disproportionately affect your regression model.