Friedman Test: Non-Parametric Repeated Measures ANOVA
What you'll learn: How to compare three or more related measurements when your data violates ANOVA's normality assumptions.
When Parametric ANOVA Won't Work for Repeated Measures
You've learned about one-way ANOVA for independent groups and the Kruskal-Wallis test as its non-parametric alternative. But what if your groups aren't independent—what if you're measuring the same subjects under multiple conditions or time points?
When you have three or more related measurements (repeated measures, matched groups, or within-subjects designs) and your data is skewed, ordinal, or violates normality assumptions, you need the Friedman test.
How the Friedman Test Works
Think of rating three different coffee brands on a 1-10 scale. You recruit 20 tasters, and each person rates all three brands. The measurements are related (same people), and rating scales aren't truly normally distributed.
The Friedman test works by:
- Ranking each subject's scores across conditions (ranks 1, 2, 3, etc.)
- Computing the sum of ranks for each condition
- Testing whether these rank sums differ more than expected by chance
It's essentially asking: "Do people systematically rank one condition higher than others?"
Key Characteristics
- Use case: Three or more related groups (repeated measures)
- Data type: Ordinal or continuous (but non-normal)
- Null hypothesis: All conditions have identical distributions
- Output: A chi-squared-like test statistic with degrees of freedom = (k - 1), where k = number of conditions
Key Takeaway: The Friedman test is your go-to non-parametric alternative for repeated measures ANOVA when normality assumptions fail—it ranks observations within each subject to detect systematic differences across conditions.