In this post, we perform a short numerical simulation to compare the statistical power of the Mann-Whitney U test and the Brunner-Munzel test under normality for various sample sizes and significance levels.
Simulation design
We conduct a simulation according to the following scheme:
- Enumerate various pairs of the significance level $\alpha$ and the sample size $n$
- Enumerate various effect sizes $ES$ from $0.1$ to $2.0$
- For each combination of the above parameters, we generate $50\,000$ pairs of random samples of size $n$: one from $\mathcal{N}(0, 1)$ and one from $\mathcal{N}(ES, 1)$. For each pair, we perform both statistical tests (one-tailed) and get the p-value. Next, we calculate the statistical power for each test based on the given value of $\alpha$
Simulation results
Here are the results for some values of $\alpha$ and $n$:
As we can see, in the presented simulations, the Brunner-Munzel test has higher statistical power than the Mann-Whitney U test (especially for small $\alpha$ and small $n$). However, it’s just a single simulation, so we can’t derive a generic conclusion about which test is better. In future posts, I will explore the behavior of these tests in more contexts.