When data isn't Normal or sample sizes are tiny, standard tests fail. We use methods based on ranks (like Mann-Whitney or Kruskal-Wallis) to draw valid conclusions without assuming a specific distribution shape (Normality).