Refer to the exhibit. These graphs were created using the GLM procedure with the plots(only)=diagnostics option. Which plot do you use to identify influential observations?

Question

Options

• ACook's D by Observation
• BResidual by Quantile
• CResidual by Predicted
• DFit - Mean and Residual Plot

• A
  82% (40)
• B
  10% (5)
• C
  6% (3)
• D
  2% (1)

Explanation

Cook's D measures how much all fitted values change when a single observation is removed, making it the standard diagnostic for identifying influential observations - those that disproportionately affect model estimates. A high Cook's D value flags an observation that, if removed, would substantially shift the regression results.

• B (Residual by Quantile) is a normality check (Q-Q plot), used to assess whether residuals follow a normal distribution - not influence.
• C (Residual by Predicted) detects heteroscedasticity and non-linearity in the error structure - useful for model fit assumptions, not influence.
• D (Fit-Mean and Residual Plot) visualizes how well the model fits across observations but doesn't quantify individual leverage or influence.

Memory tip: Think "D for Damage" - Cook's D measures how much damage one data point can do to your entire model. If a point has high Cook's D, it's doing outsized damage to your estimates.

No community discussion yet for this question.