Mann‑Whitney

The Mann‑Whitney test—also known as the Mann‑Whitney U test or Wilcoxon rank‑sum test—is one of the most practical tools in the Analyze phase when your data refuses to behave nicely. Many real-world processes produce skewed, bounded, or ordinal data that violates the assumptions of a two‑sample t‑test. Cycle times, wait times, satisfaction ratings, and defect counts often fall into this category. When normality is questionable or when outliers distort the mean, the Mann‑Whitney test provides a reliable, assumption‑light way to compare two independent groups. 

At its core, the Mann‑Whitney test evaluates whether one group tends to have higher or lower values than the other. Instead of comparing means, it compares distributions by ranking all observations together and examining how the ranks are distributed between groups. This makes the test resistant to outliers and effective even when the underlying data is heavily skewed. 

The test is especially useful when dealing with ordinal data—such as survey responses on a 1–5 scale—where calculating means is not appropriate. It also works well when sample sizes are small, provided the data is independent and the shapes of the distributions are reasonably similar. 

A common misconception is that the Mann‑Whitney test compares medians. While it is often interpreted that way, the test technically evaluates whether the probability that a randomly selected value from one group exceeds a randomly selected value from the other is different from 0.5. In practice, when distributions have similar shapes, this aligns closely with comparing medians. 

The process is straightforward. You combine the data from both groups, rank all values from lowest to highest, and calculate the sum of ranks for each group. The test statistic (U) reflects how far the observed rank distribution deviates from what would be expected if the groups were identical. A small U value indicates that one group consistently has lower ranks, while a large U value indicates the opposite. 

Interpreting the results requires attention to both statistical and practical significance. A statistically significant result indicates that the groups differ in distribution, but it does not quantify the size of the difference. Effect size measures—such as the rank‑biserial correlation—help communicate the magnitude of the difference in a meaningful way. 

In the Analyze phase, the Mann‑Whitney test is particularly valuable when comparing performance across shifts, machines, suppliers, or methods where the data is not normally distributed. It allows you to make confident, defensible decisions without forcing the data into assumptions it does not meet. When used thoughtfully, it becomes a powerful tool for uncovering meaningful differences in real-world processes.