The F-Test for Model Comparison
What you'll learn: How to use the F-test to determine whether adding predictors to a nested model significantly improves your regression.
Why Compare Models?
When building a multiple regression model, you'll often wonder: "Does adding these extra variables actually help, or am I just adding noise?" The F-test for model comparison gives you a statistical answer.
This test specifically works with nested models—where one model (the "reduced" model) is a simpler version of another (the "full" model). Think of it like comparing a basic sandwich to one with extra toppings: does the added complexity justify the extra ingredients?
How the F-Test Works
The F-test compares two things:
- How much better the full model fits the data (lower RSS—residual sum of squares)
- How many extra parameters you added (degrees of freedom cost)
The test asks: "Is the improvement in fit large enough to justify the added complexity, or could this improvement have happened by random chance?"
If the F-statistic is large and the p-value is small (typically < 0.05), the additional predictors significantly improve model fit. If not, stick with the simpler model.
Practical Example
Suppose you have a reduced model predicting salary from years of experience, and a full model that adds education level and location. The F-test tells you whether those two additional predictors are worth keeping—not just whether they slightly improve R-squared, but whether they add statistically significant explanatory power.
This directly extends the partial F-test concept you learned earlier, now applied systematically to compare any two nested models.
Key Takeaway: The F-test for model comparison tells you whether additional predictors in a nested model provide a statistically significant improvement over a simpler model, helping you balance fit against complexity.