What Is the Research Hypothesis When Using ANOVA Procedures?

In one-way ANOVA the research hypothesis is a claim about population means across three or more groups. The null hypothesis says all the group means are equal; the alternative says that at least two of them differ. Nothing stronger is claimed, and the test cannot say which means differ, only that they are not all the same.

What is the research hypothesis when using ANOVA procedures?

The research hypothesis in ANOVA is the alternative hypothesis, H_a. For a single factor with k groups, OpenStax Introductory Statistics 2e writes the pair as H₀: μ₁ = μ₂ = μ₃ = … = μ_k and H_a: at least two of the group means μ₁, μ₂, …, μ_k are not equal. The NIST/SEMATECH e-Handbook of Statistical Methods states the same idea in words: the null is that there is no difference in the population means of the different levels of the factor, and the alternative is that the means are not the same.

This makes the ANOVA research hypothesis an omnibus hypothesis. It is a single yes/no question about the whole set of groups. Your substantive theory (for example, “the three teaching methods lead to different test scores”) is usually more specific than that, and the specific claims about which groups differ, and in which direction, are tested in a second step with post-hoc comparisons or planned contrasts.

Null vs alternative hypothesis for one-way ANOVA

Element	Null hypothesis (H0)	Alternative / research hypothesis (Ha)
Symbols	μ1 = μ2 = … = μk	At least two of μ1, …, μk are not equal
In words	No difference in the population means of the factor levels	The population means are not all the same
Decision	Retained when F is small and p is large	Supported when F is large and p is below α

For the general hypothesis-testing logic, including why a test rejects or fails to reject the null rather than proving either hypothesis, see the comparison null hypothesis vs alternative hypothesis.

Why the alternative is never “all means differ”

H_a is satisfied if even one group mean departs from the others. Two groups can have identical means while a third differs, and the alternative is still true. Writing “all means are different” overstates what a significant F establishes. Likewise, a significant F does not tell you that group 3 is higher than group 1; it only says the equal-means pattern is not supported.

Why ANOVA has no one-tailed research hypothesis

With three or more groups there is no single direction to predict, so the alternative is stated as “not all equal”. The F test compares the variability between group means with the variability within groups, and a large ratio signals that the group means are more spread out than chance within-group variation would produce. A directional claim such as “treatment C exceeds treatment A” is a contrast or a post-hoc comparison, not the ANOVA alternative.

The test statistic behind the hypotheses

The one-way F statistic is the ratio of two mean squares:

F = MS_between / MS_within, where MS_between = SS_between / (k − 1) and MS_within = SS_within / (N − k).

• SS_between = Σ n_j(x̄_j − x̄)²: squared distance of each group mean x̄_j from the grand mean x̄, weighted by group size n_j.
• SS_within = Σ Σ (x_ij − x̄_j)²: squared distance of each observation from its own group mean.
• k = number of groups; N = total number of observations.

NIST notes the total sum of squares splits as SS(Total) = SST + SSE, the treatment and error components. If H₀ is true, the two mean squares estimate the same variance and F is near 1. See how to interpret the F-statistic and the F table for critical values.

Worked example with the arithmetic shown

Suppose a study compares a measured outcome score across three groups of five participants each (the numbers are made up to keep the arithmetic easy to check). The hypotheses are H₀: μ_A = μ_B = μ_C and H_a: at least two of the three means differ, with α = .05.

Group	Scores	Mean	Sum of squared deviations from group mean
A	4, 5, 6, 5, 5	5	1 + 0 + 1 + 0 + 0 = 2
B	6, 7, 7, 8, 7	7	1 + 0 + 0 + 1 + 0 = 2
C	8, 9, 10, 9, 9	9	1 + 0 + 1 + 0 + 0 = 2

1. Grand mean = (5 + 7 + 9) / 3 = 7 (groups are equal in size).
2. SS_between = 5(5 − 7)² + 5(7 − 7)² + 5(9 − 7)² = 20 + 0 + 20 = 40.
3. SS_within = 2 + 2 + 2 = 6. SS_total = 40 + 6 = 46.
4. df_between = k − 1 = 2; df_within = N − k = 15 − 3 = 12.
5. MS_between = 40 / 2 = 20; MS_within = 6 / 12 = 0.5.
6. F = 20 / 0.5 = 40.

The critical value of F with (2, 12) degrees of freedom at α = .05 is about 3.89. Because 40 is far larger than 3.89, the exact p-value is about 5 × 10⁻⁶ and we reject H₀. The data support the alternative: at least two group means differ.

Interpretation: what a significant or non-significant F means

Result	What you can say	What you cannot say
p < α (reject H0)	The population means are not all equal; at least two differ	That every group differs, which groups differ, or in which direction
p ≥ α (fail to reject H0)	The data do not provide evidence against equal means	That the means are proven equal

After a significant F, identify the specific differences with a post-hoc procedure or planned contrasts; the choice is covered in Bonferroni vs Tukey post-hoc correction. Report an effect size alongside the p-value, because the example above has a very small p-value but the p-value alone says nothing about the size of the differences. In the example, η² = SS_between / SS_total = 40 / 46 ≈ .87.

Assumptions and common mistakes

OpenStax lists the conditions for one-way ANOVA: each population from which a sample is taken is assumed to be normal; all samples are randomly selected and independent; the populations are assumed to have equal standard deviations (or variances); the factor is a categorical variable; and the response is a numerical variable. NIST likewise states the test under the assumption that the sampled populations are normally distributed.

• Stating H_a as “all means differ”. Use “at least two means differ”.
• Using symbols for sample means. The hypotheses are about population means (μ), not sample means (x̄).
• Writing “accept the null”. Say “fail to reject”.
• Running ANOVA on two groups and calling it a directional test. With two groups the F test is equivalent to a two-sided t test; for a directional question use a t test.
• Skipping the equal-variance check. See Levene’s test for equality of variances.
• Stopping at the omnibus result when the research question is about specific groups.

How to report it (APA example)

State the hypotheses in the introduction or method section, then report the test with the degrees of freedom, the F value, the exact p or p < .001, and an effect size. For the example above:

A one-way ANOVA tested the null hypothesis that mean outcome scores were equal across Groups A, B and C. The groups differed significantly, F(2, 12) = 40.00, p < .001, η² = .87.

How to calculate it in Excel, R or Python

Excel

Put each group in its own column, then use Data > Data Analysis > Anova: Single Factor (Analysis ToolPak). To get the p-value from F yourself, use:

=F.DIST.RT(40, 2, 12)

R

score <- c(4,5,6,5,5, 6,7,7,8,7, 8,9,10,9,9)
group <- factor(rep(c("A","B","C"), each = 5))
summary(aov(score ~ group))

More options are in running ANOVA in R.

Python

from scipy import stats
a = [4,5,6,5,5]; b = [6,7,7,8,7]; c = [8,9,10,9,9]
print(stats.f_oneway(a, b, c))

For a model-formula approach with a full ANOVA table, see ANOVA in Python with statsmodels. For the general definition of the method, see the dictionary entry ANOVA (Analysis of Variance).

FAQ

What is the research hypothesis when using ANOVA procedures?

It is the alternative hypothesis: at least two of the population group means are not equal. It is tested against the null hypothesis that all the group means are equal.

Is the ANOVA research hypothesis directional?

No. The F test evaluates whether the means are all equal, not which is larger. Directional predictions are examined with contrasts or post-hoc tests.

Do I write one hypothesis or several?

One omnibus pair (H₀ and H_a) for each factor in the model. Follow-up comparisons can each have their own hypotheses if they were planned.

What if my theory predicts a specific ordering of groups?

State the ordering as a planned contrast or follow-up comparison in addition to the omnibus hypothesis, and name it in your analysis plan before you see the data.

Sources

• OpenStax, Introductory Statistics 2e, section 13.1 One-Way ANOVA: https://openstax.org/books/introductory-statistics-2e/pages/13-1-one-way-anova (checked 2026-10-29).
• NIST/SEMATECH e-Handbook of Statistical Methods, section 7.4.3 Are the means equal?: https://www.itl.nist.gov/div898/handbook/prc/section4/prc43.htm (checked 2026-10-29).