What Are Residuals and Why Plot Them?
You'll learn why visualizing residuals is your best diagnostic tool for detecting regression model problems.
Understanding Residuals (Quick Recap)
You've already learned that residuals are the difference between what your model predicted and what actually happened:
residual = observed value - predicted value
They represent the part of the data your model couldn't explain—the "leftover" error.
Why Plot Them?
Here's the key insight: patterns in residuals reveal invisible problems that summary statistics like R-squared can't detect.
Think of residuals like the debris left after cleaning a room. If the debris is randomly scattered (just dust), your cleaning was uniform. But if you see clumps of dirt in corners or specific patterns, something systematic went wrong—you missed spots or your method failed in certain areas.
What Residual Plots Reveal
When you plot residuals (typically residuals vs. fitted values), you're looking for:
- Random scatter: Good! Your model assumptions hold
- Curved patterns: Your relationship isn't actually linear (violates linearity assumption)
- Funnel shapes: Variance changes as predictions increase (violates homoscedasticity)
- Outliers: Extreme values that might distort your model
- Clusters or trends: Independence assumption might be violated
A good model produces residuals that look like random noise—no patterns, just chaos. Patterns mean your model is systematically getting something wrong in certain conditions.
Key Takeaway: Residual plots transform invisible model violations into visible patterns—they're your early warning system that catches problems before you make bad predictions or wrong conclusions.