Friedman Test Explained: How to Make, Read and Report It
The Friedman test compares three or more related conditions, for instance the same participants measured under several treatments, using within-participant ranks. It is the rank-based alternative to one-way repeated-measures ANOVA.
What it tests
It tests whether conditions differ in their rank distributions within participants. Assumptions: a complete block design (each participant has a value in every condition), independence between participants, and ordinal or continuous data.
Calculation
χ²F = [12 / (n k (k + 1))] Σ Rj2 − 3n(k + 1) W = χ²F / [n(k − 1)]
Here n is the number of participants, k the number of conditions and Rj the sum of ranks for condition j. W is Kendall's coefficient of concordance, used as an effect size.
Original example (illustrative)
Task completion times (seconds) for six users on three interfaces (I1, I2, I3):
| User | I1 | I2 | I3 | Ranks |
|---|---|---|---|---|
| 1 | 12 | 14 | 19 | 1, 2, 3 |
| 2 | 11 | 16 | 15 | 1, 3, 2 |
| 3 | 13 | 15 | 18 | 1, 2, 3 |
| 4 | 15 | 14 | 20 | 2, 1, 3 |
| 5 | 10 | 13 | 17 | 1, 2, 3 |
| 6 | 12 | 15 | 19 | 1, 2, 3 |
Rank sums are R1 = 7, R2 = 12, R3 = 17. Check: 7 + 12 + 17 = 36 = 6 × 3 × 4 / 2. Then χ²F = (12/72) × (49 + 144 + 289) − 72 = 80.33 − 72 = 8.33, df = 2, p = e−8.33/2 = .016. Kendall's W = 8.33 / (6 × 2) = .69, a large degree of agreement that I3 is slowest and I1 fastest. With only six users the chi-square approximation is approximate, and an exact p-value is preferable when available.
Post hoc comparisons
Follow a significant result with pairwise Wilcoxon signed-rank tests or Nemenyi-type comparisons, adjusting for multiple testing.
How to make it in StatFAS
Open the Friedman Test Calculator. Use one row per participant and one column per condition.
APA 7 reporting
A Friedman test indicated that completion time differed across interfaces, χ²(2, N = 6) = 8.33, p = .016, Kendall's W = .69.
Common mistakes and alternatives
Using it with independent groups (use Kruskal-Wallis), dropping participants with missing values without reporting it, skipping post hoc correction, and ignoring order effects in the design; counterbalance conditions. For two conditions use the Wilcoxon test.
Conclusion
The Friedman test is the standard rank-based option for repeated measures with three or more conditions, and Kendall's W conveys how consistent the ordering is across participants.
1.
statfas-friedman-test-interface.png, alt: "StatFAS Friedman Test Calculator interface"2.
statfas-friedman-test-example-data.png, alt: "Example data entered in the StatFAS Friedman Test Calculator"3.
statfas-friedman-test-settings.png, alt: "Friedman Test Calculator settings in StatFAS"4.
statfas-friedman-test-result.png, alt: "Friedman Test Calculator result generated in StatFAS"References
- Friedman, M. (1937). The use of ranks to avoid the assumption of normality implicit in the analysis of variance. Journal of the American Statistical Association, 32(200), 675–701.