Interpreting Kruskal-Wallis Results and Effect Size
What you'll learn: How to read and interpret the H-statistic, p-values, and epsilon-squared effect size from a Kruskal-Wallis test.
Understanding the H-Statistic
The Kruskal-Wallis H-statistic measures how much the rank distributions differ across groups. Think of it like the F-statistic in ANOVA, but for ranks instead of means. A larger H-statistic suggests greater differences between groups. The H-statistic follows a chi-squared distribution under the null hypothesis, which lets us calculate p-values.
Interpreting P-Values
Just like with parametric tests, the p-value tells you the probability of seeing your results (or more extreme) if all groups truly have identical distributions. A small p-value (typically < 0.05) suggests at least one group differs from the others. However, the test doesn't tell you which groups differ—you'll need post-hoc comparisons for that.
Epsilon-Squared Effect Size
Since the Kruskal-Wallis test is non-parametric, you can't use eta-squared like in ANOVA. Instead, use epsilon-squared (ε²), which estimates the proportion of variance in ranks explained by group membership.
The formula is: ε² = H / (n² - 1) / (n + 1)
where H is the test statistic and n is total sample size.
Interpretation guidelines:
- ε² ≈ 0.01: small effect
- ε² ≈ 0.06: medium effect
- ε² ≈ 0.14: large effect
These mirror Cohen's benchmarks for effect sizes and help you understand practical significance beyond just statistical significance.
Putting It Together
When you get H = 12.5 with p = 0.002 and ε² = 0.08, you'd conclude: "There's strong statistical evidence that at least one group differs (p = 0.002), and the group membership explains about 8% of the variance in ranks—a medium-sized effect."
Key Takeaway: The H-statistic and p-value tell you if groups differ significantly, while epsilon-squared quantifies how much group membership matters practically—always report both for complete interpretation.