Back to blog
·3 min read

Friedman test step by step: when to use it, how to compute it and how to report it

The Friedman test answers one specific question: does anything change across three or more measurements of the same people? For example, a group’s satisfaction at the start, middle and end of a course, or the rating each customer gives three versions of a product.

When to use it

  • You have three or more measurements (or conditions), all answered by the same person.
  • The data aren’t normal, or they’re ordinal, like a single Likert item.
  • Each person has all their measurements: if one is missing, that person drops out of the analysis.

If the measurements are normal, the more powerful alternative is repeated-measures ANOVA. If you only have two measurements, use Wilcoxon. If they’re groups of different people, use Kruskal-Wallis.

How it’s computed, with an example

Four people rate their satisfaction at the start, middle and end of a program.

  1. In each row, rank that person’s scores: 1 for the lowest, 3 for the highest. If someone gave 3, 4 and 5, their ranks are 1, 2 and 3. Ties get the average rank.
  2. Add up each column’s ranks. Suppose you get 5 (start), 8 (middle) and 11 (end).
  3. Apply the formula: χ² = 12 / (n k (k + 1)) × Σ R² − 3 n (k + 1), with n = 4 people and k = 3 measurements. That gives 12 / 48 × (25 + 64 + 121) − 48 = 4.5.
  4. Look up the p-value in the chi-square distribution with k − 1 = 2 degrees of freedom: p = .105. With only four people, the difference doesn’t reach significance.

The mean ranks (5/4 = 1.25, 2.00, 2.75) show the direction: satisfaction rises through the program.

Pairwise comparisons

If Friedman is significant, you know at least one measurement differs, but not which. Pairwise comparisons answer that: they compare each pair of measurements and correct the p-value with Bonferroni, multiplying it by the number of comparisons. SPSS shows them in the model view under “Pairwise Comparisons”, with the “Adj. Sig.”.

Effect size: Kendall’s W

It’s computed as W = χ² / (n (k − 1)) and ranges from 0 to 1. Read it with the cutoffs 0.1 small, 0.3 medium and 0.5 large. In the example, W = 4.5 / (4 × 2) = 0.56: a large effect, even though the sample is too small for it to be significant.

How to report it in APA 7

Report each measurement’s median or mean rank, the chi-square with its degrees of freedom and sample size, the p-value and W. For example:

A Friedman test showed significant differences in satisfaction across the three measurements, χ²(2, n = 10) = 13.61, p = .001, W = .68. Bonferroni-corrected pairwise comparisons indicated that satisfaction at the end (Mdn = 4) was higher than at the start (Mdn = 3), p = .004.

Do it without SPSS

Paste your data with one row per person and one column per measurement: you get the chi-square, p-value, mean ranks, pairwise comparisons, Kendall’s W, an APA table in Word and a sentence for your thesis. Results match SPSS.