Common questions
Short answers to situations that the test chooser does not decide on its own.
Some patients appear in both groups and some in only one. Is that paired or unpaired?
Both. Compare the patients who appear in both groups with McNemar's test, and the rest with a chi-square test. Hybrid methods analyse all patients at once; ask a statistician.
My data have zeros, so I can't take the log. What now?
Add a constant to every value before transforming, usually log(x + 1). If the transformed data still aren't roughly normal, use a non-parametric test on the original data.
A t-test on log data and a Mann–Whitney give slightly different p-values. Which do I report?
The one you planned. Small differences in p are irrelevant. The t-test on the log scale has a little more power when its assumptions hold and gives an interpretable ratio; Mann–Whitney needs fewer assumptions. Stay consistent with the rest of your analyses.
A starting point, not a substitute for a statistician. See advanced topics.