What Is the F-Test and F-Distribution in Statistics?

The F-test and F-distribution are the engine behind some of the most widely used tools in inferential statistics. When a researcher tests whether three or more group means differ, they compute an F statistic. When a data analyst checks whether a linear regression model explains meaningful variation, they read an F ratio. Understanding the F-test and F-distribution — what they are, when to use them, and how to compute the F statistic — gives you the foundation for ANOVA, regression significance testing, and variance comparison.

What Is the F-Distribution?

The F-distribution (formally the Fisher-Snedecor distribution, named after Ronald Fisher and George W. Snedecor) is a continuous probability distribution defined for all positive real values. It arises naturally whenever you compute the ratio of two independent chi-square random variables, each divided by its own degrees of freedom.

If χ₁² has d₁ degrees of freedom and χ₂² has d₂ degrees of freedom, and the two are independent, then the random variable

F = (χ₁² / d₁) / (χ₂² / d₂)

follows an F-distribution with parameters d₁ and d₂, written F(d₁, d₂).

Because variance estimates computed from normally distributed data follow chi-square distributions, the F-distribution is the natural reference for comparing two variance estimates against each other. That is exactly what an F-test does.

The NIST/SEMATECH e-Handbook of Statistical Methods, Section 1.3.6.6.5 — F Distribution provides the formal probability density function and cumulative distribution tables for reference.

Shape and Properties of the F-Distribution

The F-distribution is right-skewed and bounded below by zero — an F value can never be negative, because it is a ratio of two non-negative quantities. Its shape depends on both degrees-of-freedom parameters:

When d₁ and d₂ are small, the distribution is heavily right-skewed with a long right tail.
As d₁ and d₂ grow larger, the distribution becomes less skewed and more symmetric, eventually approaching a normal distribution in the limit.

Key properties for F(d₁, d₂):

Mean: d₂ / (d₂ − 2) for d₂ > 2 (undefined when d₂ ≤ 2)
Mode: ((d₁ − 2) / d₁) × (d₂ / (d₂ + 2)) for d₁ > 2
Range: (0, ∞) — positive values only
Shape: right-skewed for all finite parameter values

Because the distribution is not symmetric, F-tables list critical values separately for each combination of (d₁, d₂) at each significance level. This is different from the t-table, where a single degrees-of-freedom column gives you the two-sided threshold for any α.

The F-Distribution and the Chi-Square Distribution

The F-distribution is a direct descendant of the chi-square distribution. If you take a chi-square variable with d₁ degrees of freedom in the numerator and a chi-square variable with d₂ degrees of freedom in the denominator, divide each by its degrees of freedom, and form the ratio, you get an F variable. A special case: an F(1, d₂) distribution is the square of a t(d₂) distribution, so F = t² when df₁ = 1. This means a one-way ANOVA comparing exactly two groups produces an F statistic equal to the square of the two-sample t statistic.

What Is an F-Test?

An F-test is any statistical hypothesis test that uses the F-distribution as its reference distribution. The core idea is consistent across all applications: compute a ratio of two variance estimates, compare that ratio to what you would expect under the null hypothesis, and decide whether the result is too extreme to attribute to chance.

The computed ratio is called the F statistic (or F ratio). A large F statistic means the variance in the numerator is much larger than the variance in the denominator. Depending on the setup, this indicates either that group means differ meaningfully or that one source of variation dominates another.

The Null Hypothesis in an F-Test

In most F-tests, the null hypothesis (H₀) asserts that the two variance estimates in the ratio come from the same underlying population variance, or equivalently that all group means are equal. Under H₀, the F statistic should hover near 1.0, because both estimates measure the same quantity and their ratio should be close to one.

A large F value — one far out in the right tail of the F-distribution — is evidence against H₀. You compare the observed F to the critical value from the F-distribution at your chosen significance level α, or you compute the exact p-value.

The Three Main Uses of F-Tests

One-way and multi-way ANOVA. Test whether the means of three or more groups are all equal by comparing the between-group variance (how much group means scatter around the grand mean) to the within-group variance (how much individual observations scatter around their own group mean).
Overall regression model fit. Test whether a regression model as a whole explains a statistically significant portion of the variability in the outcome variable. The overall F-test in linear regression checks whether R² is greater than zero in the population.
Equality of two variances. Test whether two populations have the same variance (using Bartlett’s test, Levene’s test, or the classic F-ratio of two sample variances) — often as a prerequisite check before a two-sample t-test that assumes equal variances.

The F-Statistic Formula

The general form of the F statistic is:

F = MS₁ / MS₂

where MS₁ and MS₂ are two mean squares — estimates of variance computed from the data. “Mean square” means a sum of squared deviations divided by its degrees of freedom.

F-Statistic in One-Way ANOVA

In a one-way ANOVA with k groups and N total observations, the F statistic is:

F = MS_between / MS_within

MS_between = SS_between / (k − 1)
MS_within  = SS_within  / (N − k)

where:

SS_between (between-group sum of squares) measures how much the group means deviate from the grand mean.
SS_within (within-group sum of squares, also called the error sum of squares) measures how much individual observations deviate from their own group mean.
k − 1 is the between-group degrees of freedom — this becomes df₁ for the F-distribution.
N − k is the within-group degrees of freedom — this becomes df₂.

The sums of squares are computed as:

SS_between = Σ nᵢ × (x̄ᵢ − x̄)²
SS_within  = Σ Σ (xᵢⱼ − x̄ᵢ)²

where nᵢ is the size of group i, x̄ᵢ is the group mean, x̄ is the grand mean, and xᵢⱼ is the j-th observation in group i.

F-Statistic for Comparing Two Variances

To test whether two populations have equal variances, compute the ratio of their sample variances:

F = s₁² / s₂²

By convention, place the larger variance in the numerator so that F ≥ 1. This statistic follows an F-distribution with (n₁ − 1) numerator degrees of freedom and (n₂ − 1) denominator degrees of freedom.

Worked Example: One-Way ANOVA F-Test

Suppose a researcher tests three study methods (A, B, C) on 9 students — three students per method — measuring scores on a 20-point quiz.

Data:

Student	Method A	Method B	Method C
1	6	10	2
2	8	12	4
3	10	14	6
Mean	8	12	4

Step 1 — Compute the grand mean.

x̄ = (6 + 8 + 10 + 10 + 12 + 14 + 2 + 4 + 6) / 9
  = 72 / 9
  = 8

Step 2 — Compute SS_between (between-group sum of squares).

Each group has nᵢ = 3 observations:

SS_between = 3×(8 − 8)² + 3×(12 − 8)² + 3×(4 − 8)²
           = 3×0  +  3×16  +  3×16
           = 0 + 48 + 48
           = 96

Step 3 — Compute SS_within (within-group sum of squares).

Sum the squared deviations from each group mean:

Method A: (6−8)² + (8−8)² + (10−8)² = 4 + 0 + 4 = 8
Method B: (10−12)² + (12−12)² + (14−12)² = 4 + 0 + 4 = 8
Method C: (2−4)² + (4−4)² + (6−4)² = 4 + 0 + 4 = 8

SS_within = 8 + 8 + 8 = 24

Step 4 — Compute the mean squares.

k = 3 groups,  N = 9 observations

df_between = k − 1 = 2
df_within  = N − k = 6

MS_between = SS_between / df_between = 96 / 2 = 48
MS_within  = SS_within  / df_within  = 24 / 6 = 4

Step 5 — Compute the F statistic.

F = MS_between / MS_within = 48 / 4 = 12

Step 6 — Make a decision.

With df₁ = 2 and df₂ = 6 at α = 0.05, the critical F value from standard F-tables is 5.14. Because F = 12 exceeds 5.14, we reject the null hypothesis. The data provide strong evidence that at least one study method produces a different mean quiz score than the others.

To put this in p-value terms: the exact p-value for F = 12 with df₁ = 2, df₂ = 6 is approximately 0.008 — well below the conventional α = 0.05 threshold.

How to Read an F-Distribution Table

Printed F-tables list critical values — the F statistic thresholds at a chosen significance level. Most tables cover α = 0.05; some also provide α = 0.01 and α = 0.10.

To use an F-table:

Find df₁ — the numerator degrees of freedom (for ANOVA: k − 1).
Find df₂ — the denominator degrees of freedom (for ANOVA: N − k).
Locate the cell at the intersection of the df₁ column and df₂ row.
Compare your F statistic. If F exceeds the critical value, reject H₀ at that significance level.

For the worked example: df₁ = 2, df₂ = 6, α = 0.05 → critical value = 5.14. Since F = 12 > 5.14, the result is statistically significant.

Modern software (R, Python, SPSS, Excel) computes the exact p-value directly, making table lookups optional. Still, understanding how to read an F-table reinforces what the p-value represents: the area under the F-distribution curve to the right of your observed F statistic.

Assumptions of the F-Test

The F-test gives reliable results only when its assumptions hold. Violating them can produce misleading F statistics and incorrect p-values.

1. Independence. Every observation must be independent of every other. In an experiment this usually means random assignment to groups and no repeated measurements on the same subject.

2. Normality. Observations in each group should come from a normally distributed population. With 30 or more observations per group the central limit theorem makes the F-test fairly robust to mild non-normality. With smaller samples the assumption matters more.

3. Homogeneity of variance (homoscedasticity). The population variance should be approximately equal across all groups. You can check this with Levene’s test or Bartlett’s test before running ANOVA. When group variances differ substantially, Welch’s ANOVA is a more robust alternative.

4. Correct model specification. In multi-factor or mixed models you must correctly distinguish fixed factors (whose levels are chosen deliberately) from random factors (whose levels are sampled from a larger population). The F ratio is constructed differently depending on this distinction, so misspecification leads to the wrong test.

The NIST/SEMATECH e-Handbook covers assumption checking for the F-test in Section 7.4.3.1 — Assumptions for the Two-Sample F-Test, including guidance on when robust alternatives are preferable.

Common Mistakes When Using the F-Test

Mistake 1: Treating a Significant F as Identifying Which Groups Differ

A significant F statistic in one-way ANOVA tells you only that at least one group mean is different from the others — it does not pinpoint which pairs of groups differ. To identify specific differences, you need post-hoc tests such as Tukey HSD, Bonferroni correction, or the Scheffé procedure.

Mistake 2: Ignoring the Homogeneity of Variance Assumption

When group variances differ substantially — a ratio of the largest to the smallest greater than about 4 — the standard F-test becomes unreliable. Always inspect the within-group standard deviations and run a homoscedasticity check. If the assumption is violated, use Welch’s ANOVA, which adjusts the degrees of freedom to compensate.

Mistake 3: Applying the F-Test to Ordinal or Heavily Non-Normal Data with Small Samples

The F-test presupposes normally distributed populations. With small samples (fewer than 15 observations per group) and strongly non-normal or ordinal data, consider the Kruskal-Wallis test instead — it is the non-parametric analogue of one-way ANOVA and makes no normality assumption.

Mistake 4: Forgetting That F ≈ 1 Under H₀

Students sometimes expect an F near zero when the null hypothesis is true. In fact, the expected value of F is d₂ / (d₂ − 2) — slightly above 1 for most practical sample sizes. An F near 1 is consistent with H₀ because both variance estimates are measuring the same underlying quantity and their ratio should hover around one by design.

Mistake 5: One-Tailed vs Two-Tailed Confusion

The standard F-test for ANOVA is inherently one-tailed: you always look at the right tail of the F-distribution, because a between-group variance larger than the within-group variance is the only direction of evidence against H₀ in this setup. This is different from the t-test, where two-tailed tests are the default for testing mean differences. The variance equality test (F = s₁²/s₂²) can be two-tailed if you simply want to test whether variances differ rather than which is larger.

Frequently Asked Questions

What does the F statistic measure?

The F statistic measures the ratio of two variance estimates. In ANOVA, the numerator estimates variation between group means (signal) and the denominator estimates variation within groups (noise). A large F means the between-group variation is much larger than the within-group variation — strong evidence that the group means are not all equal.

What is the p-value in an F-test?

The p-value in an F-test is the probability of observing an F statistic as large as — or larger than — the one computed, given that the null hypothesis is true. A small p-value (typically < 0.05) is evidence against H₀. In the worked example, F = 12 with df₁ = 2 and df₂ = 6 produces a p-value of approximately 0.008, indicating a statistically significant result.

The F-distribution is defined as the ratio of two independent chi-square random variables, each divided by its degrees of freedom. Because variance estimates from normally distributed data follow chi-square distributions, the F-distribution is the natural reference for variance ratios. A practical consequence: an F(1, d₂) distribution is the square of a t(d₂) distribution, so a one-way ANOVA with exactly two groups produces F = t².

What is the difference between an F-test and a t-test?

A t-test compares the means of two groups and uses the t-distribution as its reference. An F-test generalises this to three or more groups (ANOVA) or directly compares two variances. When comparing exactly two group means, a one-way ANOVA produces F = t², and the conclusions are identical. Use a t-test for two groups; use ANOVA (F-test) for three or more groups compared simultaneously.

When should I use an F-test?

Use an F-test when you want to:

Compare means across three or more groups simultaneously (one-way or multi-way ANOVA).
Test whether a linear regression model as a whole explains significant variation in the outcome (the overall F-test in regression output).
Compare the variances of two independent samples directly.

If you are comparing exactly two group means, a two-sample t-test is the more direct choice. If your data are ordinal or strongly non-normal with small samples, consider a non-parametric alternative such as the Kruskal-Wallis test in place of one-way ANOVA.

Does a higher F statistic always mean a more significant result?

A higher F statistic pushes the p-value lower, but the relationship also depends on the degrees of freedom. F = 5 with df₁ = 2, df₂ = 6 has a different p-value than F = 5 with df₁ = 10, df₂ = 100. Always interpret the F statistic in the context of the associated p-value or the critical value for the specific degrees of freedom in your study.

What does it mean when F equals 1?

An F statistic of exactly 1 means the two variance estimates in the ratio are equal. In ANOVA, this would indicate that the variation between group means is no larger than the variation within groups — exactly what you would expect if all group means were equal. Values near 1 (slightly above or below due to sampling variation) support the null hypothesis. Only values substantially greater than 1, in the right tail of the F-distribution, constitute evidence against H₀.