The Residuals vs Fitted Values Plot
You'll learn how to create and interpret the single most important diagnostic plot for checking if your linear regression model meets its core assumptions.
What It Is
The residuals vs fitted values plot is your visual health check for regression. On the x-axis, you plot the fitted (predicted) values from your model. On the y-axis, you plot the corresponding residuals (actual minus predicted). Each point represents one observation from your dataset.
Why This Plot Matters
This plot helps you check two critical assumptions at once:
Linearity: If the relationship truly is linear, residuals should scatter randomly around zero with no pattern. If you see a curve or U-shape, your relationship might not be linear.
Homoscedasticity (constant variance): The vertical spread of residuals should stay roughly the same across all fitted values. If the spread fans out (like a megaphone) or narrows, you're violating the constant variance assumption.
What to Look For
- Good: Random cloud of points scattered evenly around the horizontal line at y=0, with consistent spread
- Bad (curved pattern): Suggests non-linear relationship; linear regression isn't appropriate
- Bad (funnel shape): Indicates heteroscedasticity; variance increases or decreases with fitted values
- Bad (outliers): Points far from the rest may be influential observations
Think of it like a doctor checking your heartbeat. A healthy rhythm (random scatter) is good. Patterns (cycles, trends) suggest something's wrong.
Key Takeaway: The residuals vs fitted values plot is your first-line diagnostic tool—a random, patternless scatter with constant spread confirms your model's linearity and homoscedasticity assumptions are reasonable.