Partial F-Tests for Nested Models
What you'll learn: How to formally test whether adding or removing a group of predictors significantly improves your regression model.
The Core Idea
You've learned how the global F-test tells you if your entire model is useful, and how t-tests check individual coefficients. But what if you want to test whether adding several predictors together improves your model? That's where partial F-tests come in.
A partial F-test compares two nested models—a smaller model (with fewer predictors) and a larger model (with additional predictors). You're essentially asking: "Does the extra complexity justify the improvement in fit?"
How It Works
Think of it like comparing two explanations for something:
- Reduced model: Uses only your baseline predictors (e.g., price predicted by square footage alone)
- Full model: Adds more predictors (e.g., price predicted by square footage + bedrooms + bathrooms)
The partial F-test calculates whether the reduction in residual sum of squares (RSS) from adding those extra variables is statistically significant, or just due to chance.
When to Use It
Use partial F-tests when you want to:
- Test if a group of related variables collectively matters (e.g., all demographic variables)
- Compare simpler vs more complex models systematically
- Determine variable importance beyond individual t-tests
The test accounts for the fact that adding predictors always improves R-squared—it asks whether that improvement is meaningful enough to reject the simpler model.
Key Takeaway: Partial F-tests let you rigorously compare nested models with different predictor subsets, helping you decide whether additional variables add genuine explanatory power or just noise.