When is a non-parametric test (like Mann-Whitney U) preferred?
When the data are heavily skewed or contain outliers.
When the sample size is very large (e.g., n>1000n > 1000n>1000).
When we want to estimate population parameters.
When we know the population is perfectly normal.