The F-Statistic and F-Distribution
You'll learn how the F-statistic compares group differences to random noise, and why it follows the F-distribution under the null hypothesis.
The Core Idea
In one-way ANOVA, we want to know: "Are the differences between groups bigger than we'd expect from random variation within groups?" The F-statistic answers this by forming a ratio:
F = (Between-group variance) / (Within-group variance)
Think of it like judging a talent show. If the difference between contestants is much larger than the variation in each contestant's own performances, you can confidently say there's real talent disparity. But if between-contestant differences are small compared to how inconsistent each performer is, you can't tell them apart.
Constructing the F-Statistic
From your work with between-group and within-group variation:
- Numerator: Mean Square Between (MSB) — measures how spread out the group means are
- Denominator: Mean Square Within (MSW) — measures the pooled variation inside each group
So:
F = MSB / MSW
When the null hypothesis is true (all group means equal), both numerator and denominator estimate the same population variance, so F should be near 1. When groups truly differ, MSB inflates while MSW stays stable, pushing F higher.
The F-Distribution
Under the null hypothesis, this ratio follows an F-distribution with two parameters:
- df₁ (numerator degrees of freedom) = k - 1 (number of groups minus one)
- df₂ (denominator degrees of freedom) = N - k (total observations minus number of groups)
The F-distribution is right-skewed, starts at zero, and its shape depends entirely on these two df values. Large F-values (far in the right tail) provide evidence against the null hypothesis.
Key Takeaway: The F-statistic quantifies whether between-group differences exceed within-group noise; under H₀, it follows an F-distribution determined by the degrees of freedom of numerator and denominator variances.