Mann-Whitney U Test Calculator
Reviewed by Hamza Gujjar, Algorithm Architect & Software Developer
Last updated: August 20, 2026Comparing Two Groups Without Assuming Normality
The Mann-Whitney U test (also called the Wilcoxon rank-sum test) compares two independent samples when the data can't be trusted to follow a normal distribution — skewed measurements, ordinal survey ratings, or small samples where a t-test's assumptions are shaky. Instead of comparing means, it ranks every observation from both groups together and checks whether one group's ranks systematically outrank the other's.
The Formula
U2 = n1n2 − U1
U = min(U1, U2)
z = (U − meanU) / σU
meanU = n1n2/2, σU tie-corrected
Where R1 is the sum of ranks assigned to Group 1 within the combined, sorted dataset (using average ranks for ties).
Worked Example
Group 1: 12, 15, 18, 20. Group 2: 10, 14, 16, 19.
| Quantity | Value |
|---|---|
| n1, n2 | 4, 4 |
| U | 6.0 |
| z | −0.5774 |
| Two-tailed p-value | 0.5637 |
Computed directly from the calculator's ranking and normal-approximation formula.
With such small samples and closely interleaved values, the test finds no statistically significant difference between the two groups — a p-value well above the usual 0.05 threshold.
Where This Calculation Matters
- Non-normal or ordinal outcomes — Likert-scale survey responses, reaction times, and income data are all frequently skewed enough that a t-test's normality assumption is questionable, making this test the safer default.
- Small samples — with few observations, it's hard to verify normality at all, and this test doesn't require it.
- Outlier-resistant comparisons — because it works on ranks rather than raw values, a single extreme outlier has far less influence than it would on a mean-based test.
How to Use This Calculator
- Enter Group 1 values as a comma-separated list.
- Enter Group 2 values as a comma-separated list.
- Select Calculate to get the U statistic, z-score, and two-tailed p-value.
Related Calculations
For paired (before/after) non-parametric comparisons, use the Wilcoxon Signed-Rank Calculator. When normality can be assumed, the parametric equivalent is the T-Test Calculator.
Principles of Nonparametric Statistics: The Mann-Whitney U Test
A Mann-Whitney U calculator (also known as the Wilcoxon Rank-Sum Test) performs a nonparametric statistical hypothesis test to determine whether two independent sample groups originate from identical continuous populations. In biostatistics, clinical drug trials, and behavioral economics, the Mann-Whitney U test serves as the primary distribution-free alternative to the independent two-sample Student's t-test when data is ordinal or violates normality assumptions.
The Fundamental Mann-Whitney U Formulas
U1 = n1·n2 + [ n1·( n1 + 1 ) / 2 ] - R1
U2 = n1·n2 + [ n2·( n2 + 1 ) / 2 ] - R2
Test Statistic: U = Minimum( U1, U2 ) | Check: U1 + U2 = n1·n2
Large-Sample Normal Approximation (Z-Score)
For sample sizes where n1 > 20 and n2 > 20, the distribution of U converges to a normal distribution:
Standard Deviation: σU = √[ ( n1 · n2 · ( n1 + n2 + 1 ) ) / 12 ]
Standardized Test Statistic: Z = ( U - μU ) / σU
Assumptions of the Mann-Whitney U Test
- Continuous or Ordinal Data: Dependent variable is measured on at least an ordinal scale (Likert scales, pain ratings, non-normal wait times).
- Independent Observations: Each subject belongs exclusively to Group 1 or Group 2.
- Distributional Shape Assumption: If group distribution shapes are similar, the test evaluates differences in group Medians; if shapes differ, it tests for general Stochastic Dominance.
Step-by-Step Worked Calculation Example
Example: Evaluating Clinical Recovery Times (Days) Between Drug and Placebo
Problem: Group A (Drug: n1 = 4): [ 3, 5, 7, 9 ] | Group B (Placebo: n2 = 4): [ 6, 8, 12, 14 ]. Total N = 8. Test whether the drug significantly shortens recovery time (α = 0.05).
Step 1: Pool and rank all 8 observations in ascending order:
Values: 3 (Rank 1), 5 (Rank 2), 6 (Rank 3), 7 (Rank 4), 8 (Rank 5), 9 (Rank 6), 12 (Rank 7), 14 (Rank 8).
Step 2: Sum ranks for each group:
R1 (Drug) = 1 + 2 + 4 + 6 = 13 | R2 (Placebo) = 3 + 5 + 7 + 8 = 23
Step 3: Calculate U Statistics:
U1 = ( 4 × 4 ) + [ 4(5)/2 ] - 13 = 16 + 10 - 13 = 13
U2 = ( 4 × 4 ) + [ 4(5)/2 ] - 23 = 16 + 10 - 23 = 3
U Statistic = Min( 13, 3 ) = 3.0
Step 4: Evaluate Critical Value:
For n1 = 4, n2 = 4, the two-tailed critical U at α = 0.05 is Ucrit = 1. Because calculated U = 3 > 1, we fail to reject the null hypothesis at α = 0.05.
Conclusion: Recovery time differences are not statistically significant at the 5% level with n = 4 per group.
Tie Correction Formula for Tied Ranks
When identical score values occur across sample datasets, tied ranks are assigned their average rank value. To prevent deflation of the test statistic variance, the standard deviation incorporates a Tie Correction Factor (T):
Where ti is the number of observations tied at rank group i.
Effect Size Metric for the Mann-Whitney U Test (Rank Biserial r)
In empirical publication standards (APA), reporting statistical significance must be accompanied by the standardized Effect Size (r):
- Small Effect: r = 0.10 to 0.29
- Medium Effect: r = 0.30 to 0.49
- Large Effect: r ≥ 0.50
APA Style Reporting Standards for the Mann-Whitney U Test
In peer-reviewed academic journals, results are formally reported including sample sizes, median values, interquartile ranges (IQR), the U statistic, standardized Z-score, exact p-value, and effect size r:
Statistical Power and Relative Efficiency (A.R.E.)
In asymptotic distribution theory (Hodges & Lehmann), the Asymptotic Relative Efficiency (A.R.E.) of the Mann-Whitney U test compared to the Student's t-test:
- For perfectly normal Gaussian data: A.R.E. = 3/π ≈ 95.5% (loses less than 5% statistical power compared to the t-test).
- For heavy-tailed, skewed, or contaminated distributions (Cauchy, Exponential, Log-Normal): A.R.E. exceeds 100% (often 120% to 300%+), making Mann-Whitney far more powerful than the parametric t-test.
Sample Size Determination and Power Analysis
In preclinical biomedical research, sample sizes for the Mann-Whitney U test are estimated using Noether's Formula for Nonparametric Power:
Where P(X > Y) represents the probability that an observation from the treatment group exceeds an observation from the control group (the effect size index θ).
Common Misconception: Medians vs. Mean Ranks
While frequently described as a test of medians, the Mann-Whitney U test strictly evaluates Mean Rank Differences (Stochastic Superiority).
The test accurately reflects median differences only when the two underlying population distributions exhibit identical shape and dispersion.
Directional One-Tailed Hypothesis Testing
When researchers possess a directional prior hypothesis (e.g., active new drug will strictly accelerate healing faster than inert placebo), one-tailed p-values are obtained by halving the two-tailed p-value output.