A residual measures how far each observation is from the regression line.

Residual Plots

A residual plot graphs residuals (vertical axis) against the explanatory variable $x$ or the predicted values $\hat{y}$ (horizontal axis).

Interpreting Residual Plots

Good Fit (Linear Model Appropriate)

• Points scattered randomly around the horizontal line $y = 0$
• No obvious pattern
• Roughly equal spread throughout

Curved Pattern

• Indicates the relationship is not linear
• A curved model (quadratic, exponential, etc.) may be more appropriate
• Consider transforming the data

Fan Shape (Heteroscedasticity)

• Spread of residuals increases (or decreases) as $x$ increases
• Indicates non-constant variability
• May need a transformation

Outliers in Residuals

• Points with large residuals (far from 0) are regression outliers
• These may indicate unusual observations worth investigating

Using Residual Plots to Assess Models

Standard Deviation of Residuals ($s$)

$s = \sqrt{\frac{\sum(y_i - \hat{y}_i)^2}{n - 2}}$

Interpretation: "The actual [y-values] typically differ from the values predicted by the LSRL by about $s$ [units]."

We divide by $n - 2$ because we estimated two parameters ($a$ and $b$).

Key Properties of Residuals

1. The mean of residuals is always 0: $\bar{e} = 0$
2. The residuals have no linear relationship with $x$
3. The sum of squared residuals is minimized by the LSRL

AP Tip: On the AP exam, when asked "Is a linear model appropriate?", always refer to the residual plot (not the scatterplot or $r$). A residual plot showing random scatter indicates the linear model is appropriate.