Regression Diagnostics

5.1 Residual Analysis

Key Plots

• Residuals vs. fitted: Should show random scatter (check homoscedasticity, linearity)
• QQ plot: Check normality of residuals
• Residuals vs. order: Check independence

5.2 Influence Measures

• Leverage: $h_{ii}$ — how far $x_i$ is from other $x$'s
• Cook's distance: $D_i$ — influence of $i$-th observation on all fitted values
• DFFITS: Change in fitted values when observation $i$ is deleted

Outlier Detection

• Standardized residuals $> 3$ in absolute value
• Leverage $> 2(k+1)/n$ indicates high leverage
• Cook's distance $> 1$ indicates high influence Join Discord PreviousANOVANextMulticollinearity