The Adjusted R-Squared Formula
What you'll learn: How adjusted R-squared corrects for model complexity by penalizing the addition of predictors that don't meaningfully improve fit.
Why We Need Adjusted R-Squared
You've learned that R-squared can be misleading in multiple regression because it always increases when you add more predictors—even if those predictors add nothing meaningful. Adjusted R-squared solves this by applying a penalty based on the number of predictors in your model.
The Formula
The adjusted R-squared formula is:
R²_adj = 1 - [(1 - R²) × (n - 1) / (n - p - 1)]
Where:
- R² = the regular R-squared value
- n = number of observations (data points)
- p = number of predictors (independent variables)
How It Works
The formula penalizes you for adding predictors by adjusting the ratio (n - 1) / (n - p - 1). As p increases (more predictors), this ratio gets larger, which reduces the adjusted R-squared unless the new predictor genuinely improves the model enough to compensate.
Analogy: Think of it like a cost-benefit analysis. Regular R-squared only counts benefits (explained variance), while adjusted R-squared subtracts costs (model complexity). Only predictors that "pay their way" improve the adjusted value.
Key Behavior
- Adjusted R-squared can decrease when you add unhelpful predictors
- It can even be negative for very poor models
- It's always ≤ regular R-squared
- The penalty grows as you add more predictors relative to sample size
Key Takeaway: Adjusted R-squared penalizes model complexity by accounting for the number of predictors, making it a better metric than R-squared for comparing models with different numbers of variables.