The Mann–Whitney U test compares the relative ranks of values in two independent samples. It is a non-parametric option for continuous or ordinal data, including data that are not normally distributed. section 5.2.1 section 5.4
To calculate the statistic by hand, rank the combined observations and sum the ranks within each group. Calculate a U value from each group’s rank sum, select the lower value, and compare it with the appropriate critical value in a U table. A lower U than the threshold is statistically significant. section 5.2.1
The test focuses on relative positions rather than the numerical size of differences. A significant result indicates a difference in ranks; it should not automatically be described as a difference in medians. Explain the result in terms that match the data and the decision at hand. section 5.2.1
Choose a test that matches the group design:
- Use Mann–Whitney U for two independent samples. section 5.4
- Consider the Wilcoxon signed-rank test for two dependent samples, or the Kruskal–Wallis test for two or more independent samples. section 5.4
Comparing city temperatures
A hand calculation compares six daily weather values each from New York City and Boston. The ranks yield a lower U for New York City, which is compared with a table value to assess significance. section 5.2.1
import numpy as np #1
from scipy.stats import skewnorm, kruskal, mannwhitneyu
np.random.seed(99) #2
group_a = skewnorm.rvs(a=9, scale=2.2, size=99) + 4.5 #3
group_b = skewnorm.rvs(a=11, scale=1, size=99) + 4.6
H = kruskal(group_a, group_b) #4
U = mannwhitneyu(group_a, group_b)
print(f"Kruskal-Wallis Test, H={H[0]}, p={H[1]}") #5
print(f"Mann-Whitney U-test, U={U[0]}, p={U[1]}")
This Python listing compares two simulated skewed groups using SciPy’s Mann–Whitney U and Kruskal–Wallis tests. section 5.2.1
