Statistics 2nd ed

distribution-free

Lesson 8 — Post Hoc Tests

tukey hsd
bonferroni

When ANOVA finds a significant F, we know that not all group means are equal.
But ANOVA does not tell us which groups differ.

For that, we need post hoc tests (Latin: after this).
They compare pairs of group means while controlling for the increased chance of error.


Tukey’s Honestly Significant Difference (HSD)

When to Use:

  • Equal group sizes
  • Pairwise comparisons after one-way ANOVA

Formula:
$$\text{HSD} = q \sqrt{\frac{MS_{\text{within}}}{n}}$$

In words:
$$\text{HSD} = \text{Studentized range statistic } q \times \sqrt{\frac{\text{mean square within groups}}{\text{sample size per group}}}$$

If the difference between two means ≥ HSD, they are significantly different.

Example:
3 groups, n = 10 each, $$MS_{\text{within}} = 16.7$$, critical $$q = 3.5$$.

$$\text{HSD} = 3.5 \times \sqrt{\tfrac{16.7}{10}} = 3.5 \times 1.29 = 4.52$$

So any pair of means that differ by 4.52 or more is significant.


Bonferroni Correction

When to Use:

  • Simple and conservative
  • Divide significance level by number of comparisons

Formula:
$$\alpha' = \frac{\alpha}{m}$$

In words:
$$\text{adjusted significance level} = \frac{\text{original significance level}}{\text{number of comparisons}}$$

Example: If α = 0.05 and 10 comparisons, α′ = 0.005 per test.


Scheffé Test

When to Use:

  • Unequal sample sizes
  • Most conservative post hoc test

Formula (summary):
Scheffé’s critical F = (k – 1) × F(critical, df_between, df_within).


Definition

  • Post hoc test: statistical test used after ANOVA to identify which means differ
  • Tukey HSD: balanced groups, pairwise
  • Bonferroni: adjusts α for multiple comparisons
  • Scheffé: conservative, flexible for unequal n

Visual Placeholders

Figure 8.1 — Tukey HSD example: three group means with horizontal bars showing which pairs differ.

Figure 8.2 — Bonferroni correction illustration: α = 0.05 split into smaller pieces.


Why This Matters

ANOVA tells us there is a difference somewhere.
Post hoc tests tell us where.
They protect against false positives while allowing multiple group comparisons.

Practice self-test quiz

In the space below, please find practice problems and self-test quizzes. For full access, please signup free.