Source: Comprehensive Guide to One-Way ANOVA Analysis Techniques, STAT 101 (Purdue University)
Difficulty: Intermediate | Prerequisites: Part 1 of these notes (ANOVA Foundations and Assumptions), comfort with summation notation.
This is the computational and interpretive half of one-way ANOVA. Part 1 covered what ANOVA is, why it uses variance and what assumptions it requires. Here you learn how the F-statistic is calculated from the ANOVA table, how to read a p-value from that statistic and what conclusions you can (and cannot) draw. You also see how ANOVA relates to the t-test you already know. If you are revising for an exam, this is where the calculation and interpretation questions live.
The F-statistic is the ratio of between-group mean square to within-group mean square. A large F (small p-value) leads you to reject the null hypothesis that all means are equal. The ANOVA table organises the degrees of freedom, sums of squares, mean squares and the F-value in one place. For two groups, the F-test and the t-test give the same answer.
F-statistic (F-ratio)
The test statistic for ANOVA, calculated as MSA / MSE. It measures how large the between-group differences are relative to the within-group noise. Think of it as a signal-to-noise ratio for group means.
Degrees of freedom (df)
The number of independent pieces of information that go into an estimate. In ANOVA there are three:
Total: df_T = n - 1
Between groups: df_A = k - 1
Within groups: df_E = n - k
where n is the total number of observations and k is the number of groups.
Sum of squares, total (SST)
The sum of squared deviations of every observation from the grand mean. It captures all the variability in the dataset.
Sum of squares, between groups (SSA)
The sum of squared deviations of each group mean from the grand mean, weighted by sample size. It captures variability due to differences among the groups.
Sum of squares, within groups (SSE)
The sum of squared deviations of each observation from its own group mean. It captures variability due to individual differences within groups.
Mean square between (MSA)
SSA divided by its degrees of freedom (k - 1). This is the between-group variance estimate.
Mean square within / mean square error (MSE)
SSE divided by its degrees of freedom (n - k). This is the within-group variance estimate and also serves as the pooled estimate of σ².
p-value
The probability of observing an F-statistic as extreme as (or more extreme than) the one calculated, assuming H₀ is true. A small p-value (typically < 0.05) is evidence against H₀.
F-distribution
The sampling distribution of the F-statistic under H₀. It is right-skewed, starts at zero and is defined by two parameters: the numerator degrees of freedom (k - 1) and the denominator degrees of freedom (n - k).
The F-statistic is the ratio MSA / MSE.
MSA reflects how spread out the group means are. MSE reflects how spread out the individual observations are within each group.
A large F means the group means are more spread out than you would expect from within-group noise alone, pointing to a real effect of the factor.
Under H₀ (all means equal), the F-statistic follows an F-distribution with df₁ = k - 1 and df₂ = n - k.
The F-distribution is always non-negative and right-skewed. The test is always one-tailed (right tail): you only reject H₀ for large values of F.
The ANOVA table is a standard way to lay out the calculation. Its columns are: Source, df, SS, MS and F.
Source | df | SS | MS | F |
|---|---|---|---|---|
Between groups | k - 1 | SSA | MSA = SSA / (k - 1) | MSA / MSE |
Within groups | n - k | SSE | MSE = SSE / (n - k) | |
Total | n - 1 | SST |
Key relationships to remember:
SST = SSA + SSE (the total variation splits cleanly).
df_T = df_A + df_E, i.e. (n - 1) = (k - 1) + (n - k).
Exams frequently give you some entries and ask you to fill in the rest.
Compare the p-value to your chosen significance level α (usually 0.05).
If p < α, reject H₀. Conclude that at least one group mean differs from the others.
If p ≥ α, do not reject H₀. There is insufficient evidence of differences among the means.
A non-significant result does not prove all means are equal. It may reflect low power (small samples, high variability).
A significant result does not tell you which groups differ. Post-hoc tests (Tukey's HSD, Bonferroni, Scheffé) are needed for that.
Confidence intervals for the difference between specific pairs of means are another route to identifying where the differences lie.
When k = 2 (only two groups), the F-test and the two-sample t-test give identical conclusions: F = t² and the p-values match.
For two groups, the t-test is often preferred because it is simpler and allows one-sided alternatives directly.
For three or more groups, ANOVA is the correct approach. Running multiple pairwise t-tests inflates the familywise error rate.
Classical ANOVA assumes equal variances across groups. The t-test has a Welch variant that relaxes this assumption for two groups.
The F-statistic:
F = \frac{MSA}{MSE} = \frac{SSA / (k - 1)}{SSE / (n - k)}Degrees of freedom:
df_A = k - 1 \quad \text{(between groups)}df_E = n - k \quad \text{(within groups)}df_T = n - 1 \quad \text{(total)}The partition of variation:
SST = SSA + SSERelationship between F and t for two groups:
F = t^2 \quad \text{when } k = 2In practice, you read the ANOVA table output from software (R, SPSS, Excel, Python). A quality-control engineer might use a one-way ANOVA to test whether three production lines yield the same average tensile strength. The ANOVA table gives the F-value and p-value in one line, telling the engineer whether to investigate further or leave the process alone.
"A large F-statistic always means a large practical difference." A large F can come from a tiny difference in means if the within-group variation is very small or the sample sizes are very large. Always consider effect size alongside significance.
"The p-value is the probability that H₀ is true." It is not. It is the probability of seeing data this extreme if H₀ were true. This is a general misunderstanding of p-values, not specific to ANOVA, but it comes up constantly in exam questions.
"You only need the F-value to draw conclusions." You need the F-value and its degrees of freedom to find the p-value from the F-distribution. The same F-value can be significant or not depending on df.
"ANOVA is always right-tailed." This is correct, but students sometimes try to do a two-tailed F-test by analogy with t-tests. The F-statistic is a ratio of variances and is always non-negative, so only large values are evidence against H₀.
⚠️ Be able to fill in a partially completed ANOVA table. Given any combination of n, k, SSA, SSE, SST, know how to recover the missing values using SST = SSA + SSE and the degrees-of-freedom formulas.
⚠️ Know the decision rule: compare p-value to α, or equivalently compare F to the critical value from an F-table.
⚠️ Understand why F = t² for k = 2 and be prepared to verify this numerically.
⚠️ A significant F-test tells you "at least one mean differs." It does not tell you which. Expect a question that probes this.
Fill in the blank: F = ______ / ______. MSA / MSE.
True or false: The degrees of freedom for the between-group source are n - 1. False. They are k - 1.
True or false: When comparing two groups, F = t². True.
Fill in the blank: If SST = 120 and SSA = 45, then SSE = ______. 75.
True or false: The F-distribution can produce negative F-values. False. F is always non-negative.
Q: An ANOVA table shows k = 4 groups, n = 28 total observations, SSA = 60 and SSE = 180. Complete the table and compute the F-statistic.
A: df_A = k - 1 = 3. df_E = n - k = 24. MSA = 60 / 3 = 20. MSE = 180 / 24 = 7.5. F = 20 / 7.5 = 2.67. SST = 60 + 180 = 240, df_T = 27.
Q: In the example above, suppose the critical F-value at α = 0.05 with df₁ = 3 and df₂ = 24 is 3.01. What is your conclusion?
A: F = 2.67 < 3.01, so you do not reject H₀. There is insufficient evidence at the 5% level that the group means differ.
Q: A student runs a one-way ANOVA on two groups and gets F = 4.84. Their friend runs a two-sample t-test on the same data and gets t = 2.20. Are both results consistent? Explain.
A: Yes. For two groups, F = t². Here 2.20² = 4.84, which matches. The p-values will be identical.
Q: An ANOVA yields p = 0.003. A student writes: "Group B has a significantly higher mean than Group A." What is wrong?
A: ANOVA only tells you that at least one mean differs somewhere among all the groups. It does not identify which pair of groups differs or in which direction. The student needs a post-hoc test to make that specific claim.
The ANOVA table structure reappears in regression analysis, where SSR (regression) + SSE (residual) = SST, and the F-test checks whether the model explains significant variation. Two-way ANOVA extends this framework by adding a second factor and an interaction term. Non-parametric alternatives (Kruskal-Wallis test) cover situations where the normality assumption fails.
F-statistic, F-ratio, F-test, F-distribution, ANOVA table, degrees of freedom, sum of squares, SST, SSA, SSE, mean square between, mean square within, MSA, MSE, p-value, significance level, alpha, critical value, post-hoc test, Tukey HSD, Bonferroni, t-test vs F-test, F equals t squared, one-way ANOVA calculations, interpreting ANOVA, STAT 101, intro statistics, Purdue