Course contentsShow
← Mathematics
Lesson 3088 of 3,12160. Statistics — ModelingFree lesson

Cook's Distance and Influence Measures

Calculate and interpret Cook's distance to detect observations that strongly influence regression results.

Cook's Distance and Influence Measures

You'll learn how to identify individual observations that disproportionately shape your regression model's conclusions.

What is Cook's Distance?

Not all data points are created equal in regression. Some observations have outsized influence on your fitted line—remove one point and suddenly your slope or intercept changes dramatically. Cook's Distance quantifies this influence by measuring how much all fitted values would shift if you deleted a single observation and re-ran the regression.

Think of it like this: imagine you're balancing a plank on a fulcrum. Most pebbles placed on the plank barely affect the balance. But one heavy rock at the edge? That tilts everything. Cook's Distance identifies those "rocks."

How It Works

Cook's Distance for observation i combines two ideas you've already learned:

  • Leverage (from the previous lesson): how far the predictor values are from average
  • Residual size: how poorly the model fits that point

The formula integrates both: a point is influential if it's unusual and doesn't fit well. The metric is usually denoted D_i or Cook's D.

Interpreting Cook's Distance

A common rule of thumb: investigate any observation where Cook's Distance exceeds 4/n (where n is sample size) or is substantially larger than others. Large values suggest that observation is "pulling" the regression line toward itself.

What to do? Don't automatically delete influential points—investigate why they're unusual. They might reveal important patterns, data entry errors, or special cases worthy of separate analysis.

Key Takeaway: Cook's Distance identifies observations that strongly influence regression results by combining leverage and residual magnitude; values above 4/n warrant investigation, not automatic removal.