Practical Example: Variable Selection
You'll learn: How to use adjusted R-squared to decide which predictors actually improve your multiple regression model.
The Scenario
Imagine you're predicting house prices and you've collected data on square footage, number of bedrooms, age of the house, and distance to the nearest coffee shop. Should you include all four predictors? Let's use adjusted R-squared to decide.
Step-by-Step Walkthrough
Model 1: Start with just square footage
- R² = 0.65, Adjusted R² = 0.649
Model 2: Add bedrooms
- R² = 0.71, Adjusted R² = 0.705
The adjusted R² increased, meaning bedrooms add real value beyond just adding more variables.
Model 3: Add house age
- R² = 0.73, Adjusted R² = 0.723
Still improving! Age contributes meaningful information.
Model 4: Add distance to coffee shop
- R² = 0.731, Adjusted R² = 0.719
Notice: R² went up slightly (it always does), but adjusted R² actually dropped. This tells you that coffee shop distance doesn't justify its inclusion—it's adding complexity without enough explanatory power.
The Decision
Keep square footage, bedrooms, and age. Drop the coffee shop variable. You've successfully used adjusted R-squared to balance model fit against model complexity.
Key Takeaway: When comparing models during variable selection, choose the one with the highest adjusted R-squared—it indicates you're adding predictors that genuinely improve predictions, not just inflating regular R² by chance.