Compare groups without assuming normality. Paste two columns for the Mann-Whitney U test, or three or more for Kruskal-Wallis H — get the test statistic, p-value, and group medians, all client-side.
Paste your groups as columns and press Calculate to run the rank-based test.
Watch: Chi-Square Tests and Non-Parametric Methods
Your data is skewed, so a t-test assumption of normality does not hold. Enter the lab to compare groups with a rank-based test instead.
Paste each group in its own column (different items in each group, not paired). Two groups use Mann-Whitney; three or more use Kruskal-Wallis.
All values are ranked from smallest to biggest across all groups. Tied values get the average rank. Working with ranks instead of raw values is what makes this test free of normality assumptions.
For two groups, it compares how ranks are spread between them. For three or more, it compares ranks across all groups. You get a p-value showing whether at least one group's middle value (median) genuinely differs.
Nonparametric tests compare groups using ranks rather than raw values, so they make no assumption about the shape of the distribution. The Mann-Whitney U test handles two independent groups; the Kruskal-Wallis H test extends the idea to three or more. They are the go-to choice when data is skewed, ordinal, or peppered with outliers.
Nonparametric (distribution-free) tests answer "do these groups differ?" without assuming the data follows a normal distribution. They replace each observation with its rank in the pooled data and analyse those ranks. Because ranks are unaffected by the exact shape of the distribution or by extreme values, the tests stay valid for skewed, ordinal, or outlier-heavy data where a t-test or ANOVA would be questionable.
These tests work on ranks rather than the raw numbers, which is what makes them safe for skewed data and outliers. Every value from all groups is pooled and ranked from smallest to largest, and the test checks whether one group's ranks are systematically higher or lower than another's. Mann-Whitney compares two groups; Kruskal-Wallis compares three or more. The result is a p-value: below 0.05 means the groups genuinely differ. Because they ignore the exact distances between values, no 'normal data' assumption is needed.
A team asks 12 customers to rate satisfaction before and after a process change on a 1–10 scale. The scores aren't normally distributed — most rated 8 or 9, with a few outliers at 3. A standard t-test isn't reliable here.
A Wilcoxon signed-rank test compares the before-and-after scores without assuming any particular shape. Result: p = 0.02 — the improvement is statistically real.
What to do with this: Confirm the change works and roll it out. When data is ordinal (ratings, ranks) or clearly non-normal, use a nonparametric test for an honest answer.
Real-world quality data — cycle times, waiting times, defect counts, satisfaction scores — is frequently skewed or bounded. Forcing a t-test onto such data risks misleading conclusions. Nonparametric tests give a defensible answer to "did this change make a difference?" even when the data is messy, so you can act with confidence.
Reach for a nonparametric test when a normality check fails, the sample is small, the data is ordinal (ratings, grades), or there are influential outliers you cannot justify removing. Mann-Whitney replaces the 2-sample t-test; Kruskal-Wallis replaces one-way ANOVA. Both compare the location (median) of the groups.
In the Analyse phase of DMAIC, you often compare before/after or supplier/shift groups whose data is not normal. Nonparametric tests let you screen for significant differences robustly, then quantify the practical effect with medians, supporting an evidence-based tollgate decision.
If the test is significant, report the group medians and the size of the gap alongside the p-value. For Kruskal-Wallis, follow up with pairwise Mann-Whitney tests (adjusting for multiple comparisons) to identify which groups differ. Confirm the difference is operationally meaningful, document it in your project storyboard, and monitor the improved process with a run chart or control chart. If the result is not significant, consider whether a larger sample would help, or whether the effect is simply too small to matter.
Combine nonparametric tests with run charts, control charts, and a DMAIC project structure to verify and sustain improvements on real-world data.
Use a nonparametric (rank-based) test when your data does not meet the assumptions of a parametric test — for example when the data is skewed, contains outliers, is ordinal, or the sample is too small to confirm normality. Instead of comparing means, these tests compare ranks (and effectively medians), so they are robust to non-normal distributions. The Mann-Whitney U test compares two independent groups; the Kruskal-Wallis H test extends this to three or more groups.
Both compare two independent groups, but the t-test compares means and assumes the data is approximately normal, while the Mann-Whitney U test compares the ranks of the combined data and makes no normality assumption. Use the t-test when normality holds and you care about the mean difference; use Mann-Whitney when the data is skewed, ordinal, or contains outliers. Mann-Whitney is only slightly less powerful than the t-test when normality actually holds.
Kruskal-Wallis is the nonparametric equivalent of one-way ANOVA. ANOVA compares the means of three or more groups and assumes normal data with equal variances; Kruskal-Wallis compares the rank distributions (effectively medians) and makes no normality assumption. The H statistic is compared to a chi-square distribution with k−1 degrees of freedom. A significant result means at least one group differs; follow up with pairwise Mann-Whitney tests to find which.
When two or more observations share the same value, they are each given the average of the ranks they would otherwise occupy — for example two values tied for 3rd and 4th place both receive rank 3.5. The calculator applies a tie correction to the variance of the Mann-Whitney z statistic and to the Kruskal-Wallis H statistic, so heavy ties do not distort the result. A large number of ties can reduce the test's power.
A significant rank-based test confirms your groups genuinely differ. Structured training shows you how to choose the right test and act on the result.
More calculators for statistics, quality, and improvement.
Want to choose and run the right hypothesis test inside a full DMAIC project? The Black Belt course covers advanced statistics, hypothesis testing, and the full analytical toolkit.
View Black Belt →