13.3 Facts About the F Distribution

Characteristics and Applications of the F Distribution

The F distribution lets you compare two variance estimates to see if they differ more than random chance would explain. It shows up constantly in ANOVA and regression, where you're testing whether group differences or model terms are statistically significant. Understanding its properties helps you interpret F-statistics and make correct decisions in hypothesis tests.

more resources to help you study

practice questions

Shape of the F Distribution Curve

The F distribution is always right-skewed (positively skewed). The curve starts at 0 on the x-axis, peaks close to the y-axis, then gradually tapers off to the right without ever reaching a fixed endpoint. It can only take on non-negative values, which makes sense because it's built from a ratio of variances (and variances can't be negative).

Two parameters control the shape: the numerator degrees of freedom ($df_1$) and the denominator degrees of freedom ($df_2$). As both increase, the distribution becomes less skewed and more symmetric.

A few formulas worth knowing:

• Mean: $\frac{df_2}{df_2 - 2}$, defined only when $df_2 > 2$. Notice the mean is always slightly above 1 when $df_2$ is moderate or large, which reflects the fact that under the null hypothesis, you'd expect the variance ratio to hover near 1.
• Variance: $\frac{2(df_2)^2(df_1 + df_2 - 2)}{df_1(df_2 - 2)^2(df_2 - 4)}$, defined only when $df_2 > 4$.

Effects of Degrees of Freedom

The F distribution is defined by its two degrees of freedom, and they play different roles:

• $df_1$ (numerator) comes from the variance estimate in the numerator of the F ratio. In one-way ANOVA, this equals the number of groups minus 1.
• $df_2$ (denominator) comes from the variance estimate in the denominator. In one-way ANOVA, this equals the total sample size minus the number of groups.

How the shape changes:

• With small degrees of freedom, the curve is heavily right-skewed with a long tail.
• As degrees of freedom increase, the curve becomes more symmetric and eventually approximates a normal distribution as both $df_1$ and $df_2$ approach infinity.
• Critical values decrease as degrees of freedom increase for a given significance level $\alpha$. This means with larger samples, a smaller F-statistic can still be significant.

The F Statistic: What It Is and Where It's Used

The F statistic is a variance ratio. In its general form:

$F = \frac{MS_{\text{between}}}{MS_{\text{within}}}$

where $MS$ stands for "mean square," which is a sum of squares divided by its degrees of freedom. You can also write this more generically as $F = \frac{s_1^2}{s_2^2}$, where $s_1^2$ and $s_2^2$ are two independent variance estimates already divided by their respective degrees of freedom.

The F statistic ranges from 0 to positive infinity. A value near 1 suggests the two variance estimates are similar (consistent with the null hypothesis). A value much larger than 1 suggests the numerator variance is bigger than you'd expect by chance.

Where you'll encounter it:

• One-way ANOVA compares means across levels of a single factor (e.g., comparing test scores across three teaching methods). This is the main focus of this unit.
• Two-way ANOVA extends this to two factors and their interaction (e.g., teaching method and class size).
• Regression analysis uses an F-test to determine whether the explanatory variables collectively have a significant relationship with the response variable.
• F-test for equality of variances directly compares the spread of two populations, which matters when checking assumptions for procedures like the independent samples t-test.

Hypothesis Testing with the F Distribution

The F distribution is a continuous probability distribution, so you find probabilities by looking at areas under the curve rather than at individual points.

Here's how hypothesis testing works with the F distribution:

1. State the hypotheses. The null hypothesis ($H_0$) typically claims no difference (e.g., all group means are equal in ANOVA). The alternative ($H_a$) claims at least one difference exists.
2. Calculate the F statistic from your sample data using the appropriate formula for your test.
3. Find the p-value by determining the area to the right of your F statistic under the F distribution with the correct $df_1$ and $df_2$. Because the F distribution is right-skewed and you're looking for evidence that the variance ratio is larger than expected, ANOVA F-tests are always right-tailed.
4. Compare to your significance level ($\alpha$). If the p-value is less than $\alpha$, reject $H_0$. Alternatively, compare your F statistic to the critical F-value from a table; reject $H_0$ if your statistic exceeds the critical value.

One thing that trips people up: a significant F-test in ANOVA tells you that at least one group mean differs, but it doesn't tell you which ones. You'd need follow-up procedures (like post-hoc tests) for that.