Picking a test is a decision, not a guess. Ask four questions: What is the outcome? How many groups? Independent or paired? Do the assumptions hold?
Why it matters
A wrong test gives wrong p-values and misleading conclusions, even if the arithmetic is perfect.
Key ideas
- Outcome type (numeric / categorical) decides the test family
- Number of groups: one, two, or three or more
- Independent groups vs paired measurements (same people twice)
- Parametric tests need roughly normal data or large samples
- If assumptions fail, choose the non-parametric alternative
Real-life research example
Compare blood pressure between a treatment group and a control group of different people. Numeric outcome + two independent groups → independent samples t-test.
When to use
- Use the decision tool before every analysis
- Decide the test when you plan the study, not after seeing the data
When NOT to use
- Don't choose a test because it gave p < 0.05
- Don't use an independent-groups test on paired data
Mini example
Same patients, BP before vs after treatment → paired design → paired t-test (or Wilcoxon signed-rank if the differences are clearly non-normal).
Quick takeaway
Outcome → groups → independent or paired → assumptions. Four questions, one right test.
Try it yourself
Use the decision tool below. Try this scenario: 'Compare blood pressure between two separate groups.' Read the recommendation and the 'Why?' box.
Numeric outcome + two independent groups + assumptions satisfied → Independent samples t-test. Because the outcome is continuous and the two groups are independent.