Reach for these when the outcome is ordinal, or badly skewed / outlier-ridden in a small sample. With a large n, the parametric tests are robust anyway.
1 · Check the assumptions. Every test above has them: see the Assumption Checks sheet, or §2.6. 2 · Pick one test, not many. Running several and keeping whatever hits p < .05 manufactures false positives, the garden of forking paths. Decide before you look. 3 · A p-value needs an effect size. Significant ≠ large — always report one beside it (§3.1). 4 · One numeric + one categorical variable is just a group comparison in disguise → box ①.