Nonparametric Effect Size Calculator

Measure practical differences beyond p-values using robust rank methods. Analyze tied and unequal data confidently. Get clear statistics, confidence cues, plots, exports, and explanations.

Calculator

Use commas, spaces, semicolons, or line breaks between values. For paired analysis, keep values in the same order across Sample A and Sample B.

Choose the nonparametric framework that matches your design.
Controls table values, summaries, and graph labels.
Used for approximate z and p values only.
Enter the first sample or first paired series.
Enter the second sample or second paired series.
Optional for multi-group analysis.
Leave empty when not needed.
Useful for wider rank comparisons.

Example data table

Scenario Sample A / Group A Sample B / Group B Extra groups Suggested method
Independent rank comparison 12, 14, 15, 17, 18, 19 8, 9, 11, 13, 14, 15 None Mann-Whitney U
Before versus after for the same cases 52, 50, 48, 46, 54, 57 49, 47, 46, 44, 50, 53 None Wilcoxon signed-rank
Three independent groups 10, 11, 13, 14, 15 16, 18, 17, 19, 20 Group C: 12, 13, 12, 14, 16 Kruskal-Wallis

These example values are only for demonstration. Replace them with your own observations before exporting results.

Formula used

1) Mann-Whitney U effect sizes

Use this for two independent groups. The calculator ranks all observations together, applies average ranks for ties, then derives effect sizes from the rank structure.

U₁ = R₁ − n₁(n₁ + 1) / 2 Rank-biserial correlation = [2U₁ / (n₁n₂)] − 1 Cliff's delta = (wins − losses) / (n₁n₂) Common-language effect = (wins + 0.5 × ties) / (n₁n₂) Standardized r = z / √N

2) Wilcoxon signed-rank effect sizes

Use this for paired or repeated measurements. Zero differences are removed before ranking absolute differences.

Matched rank-biserial = (W⁺ − W⁻) / [n(n + 1) / 2] Standardized r = z / √n Median paired difference = median(A − B)

3) Kruskal-Wallis effect sizes

Use this for two or more independent groups when rank-based comparison is preferred over parametric assumptions.

H = {12 / [N(N + 1)]} × Σ(Rⱼ² / nⱼ) − 3(N + 1) Tie correction: C = 1 − Σ(t³ − t) / (N³ − N) Corrected H = H / C Epsilon squared = (H − k + 1) / (N − k) Eta squared (H) = (H − k + 1) / (N − 1)

Approximate z and p values are provided as practical cues, not exact small-sample inference.

How to use this calculator

  1. Select the test family that matches your study design.
  2. Paste numeric observations into the visible fields using commas, spaces, semicolons, or line breaks.
  3. For paired analysis, keep Sample A and Sample B aligned by row or position.
  4. Choose decimal precision and decide whether to apply continuity correction.
  5. Press Calculate effect size to show results below the header and above the form.
  6. Review the metrics table, group summary, and Plotly graph.
  7. Use the export buttons to save a CSV table or PDF summary.

FAQs

1) What does a nonparametric effect size show?

It shows the practical magnitude and direction of differences without depending heavily on normality assumptions. Instead of only asking whether groups differ, it describes how strongly ranks or paired changes separate the samples.

2) When should I use Mann-Whitney mode?

Use Mann-Whitney when you have two independent groups, such as treatment versus control, and you want rank-based effect measures like rank-biserial correlation, Cliff’s delta, and common-language probability.

3) When is Wilcoxon signed-rank the correct choice?

Use Wilcoxon when the same units are measured twice or when observations are naturally paired. The calculator compares paired differences and converts the signed-rank pattern into effect size measures.

4) Why does Kruskal-Wallis report epsilon squared?

Epsilon squared summarizes how much of the rank variation is associated with group membership. It provides a practical magnitude estimate for multi-group rank comparisons and is often easier to interpret than H alone.

5) How are ties handled?

Tied observations receive average ranks. Tie groups also affect the variance correction used for approximate z values and the corrected H statistic, which helps keep the rank-based estimates more realistic.

6) Are the p values exact?

No. The calculator reports approximate p values from normal-style standardization when available. They are useful screening cues, but exact inference may require dedicated procedures for very small samples.

7) Can I paste values from spreadsheets?

Yes. You can paste comma-separated, space-separated, semicolon-separated, or line-separated values. Non-numeric tokens are ignored and listed in a warning so you can clean the data quickly.

8) Which metric should I report in practice?

Report the metric aligned with your test design: rank-biserial or Cliff’s delta for independent groups, matched rank-biserial for paired data, and epsilon squared for Kruskal-Wallis designs. Add medians and sample sizes for context.

Related Calculators

brunner munzel testspearman rank correlation calculatorkolmogorov smirnov test calculatoriqr calculatorkernel density estimatorfisher exact test calculatorgoodman kruskal gammacramer v calculatortheil sen estimatorkruskal wallis test calculator

Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.