A significant ANOVA tells you that the groups differ somewhere, not where. Post hoc tests compare the groups pair by pair while controlling the chance of a false positive across all the comparisons. The APA 7 write-up names the test, reports the adjusted p values, and makes the direction of each difference clear. This guide shows the format for Tukey's HSD, Bonferroni and Games-Howell, with a worked example.
What you need before you write a single word
- The omnibus test: F, both df, exact p and effect size from the ANOVA.
- The name of the post hoc test and why it fits (equal variances: Tukey; unequal variances: Games-Howell; a few planned comparisons: Bonferroni).
- Each group's mean (M) and SD.
- For each comparison: the mean difference, the adjusted p value, and ideally its 95% confidence interval.
The APA 7 format template
A one-way ANOVA showed a significant effect of [factor] on [outcome], F(df1, df2) = X.XX, p = .XXX, η² = .XX. Tukey's HSD post hoc tests showed that [group A] (M = X.XX, SD = X.XX) scored significantly lower than [group B] (M = X.XX, SD = X.XX), p = .XXX, 95% CI [X.XX, X.XX]. [Group B] and [group C] did not differ significantly, p = .XXX.
All 53 templates, ready to paste with the italics set: the Rescue Vault, $27 →
Formatting rules reviewers actually check:
- Name the test (Tukey's HSD, Bonferroni, Games-Howell, Holm). "Post hoc tests showed" alone is not enough.
- Report the adjusted p values, the ones your software corrected for multiple comparisons.
- Give the means so the reader can see the direction of each difference.
- For many comparisons, put them in a table and summarise the pattern in the text.
A worked example
Suppose 75 adults followed one of three diets for 12 weeks, and you compared weight loss (kg).
- Low-carb (n = 25): M = 6.07, SD = 2.32
- Low-fat (n = 25): M = 4.38, SD = 2.43
- Mediterranean (n = 25): M = 6.24, SD = 2.05
- ANOVA: F(2, 72) = 5.10, p = .009, η² = .12
- Levene's test: F(2, 72) = 0.13, p = .876, so Tukey's HSD fits
Written up in APA 7:
A one-way ANOVA showed a significant effect of diet on weight loss, F(2, 72) = 5.10, p = .009, η² = .12. Tukey's HSD tests showed that the low-fat group (M = 4.38, SD = 2.43) lost significantly less weight than both the low-carb group (M = 6.07, SD = 2.32), p = .028, 95% CI of the difference [0.15, 3.23], and the Mediterranean group (M = 6.24, SD = 2.05), p = .014, 95% CI [0.32, 3.39]. The low-carb and Mediterranean groups did not differ significantly, p = .964.
Post hoc comparisons in a table
| Comparison | Mean difference | 95% CI | p (Tukey) |
|---|---|---|---|
| Low-carb vs low-fat | 1.69 | [0.15, 3.23] | .028 |
| Mediterranean vs low-fat | 1.85 | [0.32, 3.39] | .014 |
| Mediterranean vs low-carb | 0.17 | [−1.37, 1.70] | .964 |
Note. Weight loss in kilograms over 12 weeks.
Bonferroni and Games-Howell versions
With Bonferroni-corrected t-tests on the same data, the two significant differences hold (p = .046 and p = .016) and the third does not (p > .999). Write "Bonferroni-corrected comparisons showed..." and report the corrected values. When the group variances differ, use Games-Howell instead and say so; the sentence is otherwise the same. See Welch's ANOVA for an example.
Mistakes reviewers catch
- Unadjusted p values. Running plain t-tests between every pair without a correction inflates false positives.
- Post hoc tests after a non-significant ANOVA. Without a significant omnibus result (or planned comparisons stated in advance), the pairwise tests should not be reported as findings.
- "Group A differed from group B" with no direction. Say which was higher, with the means.
- A Bonferroni p above 1. Software caps corrected p values at 1; report p > .999.
- Tukey with unequal variances. Use Games-Howell.
Let KyroStat do the write-up for you
KyroStat picks the post hoc test that fits your variances, runs every comparison, and writes the paragraph and the comparison table from your data.
Frequently asked questions
Tukey or Bonferroni? Tukey's HSD is designed for all pairwise comparisons and is usually more powerful there. Bonferroni suits a small number of planned comparisons.
Do I report every comparison? Report the ones that answer your question in the text, and all of them in a table if there are many.
What about non-parametric post hoc tests? After a Kruskal-Wallis test, use Dunn's test with a Bonferroni or Holm correction and report medians.
How do I report the ANOVA itself? See how to report a one-way ANOVA.