Which statistical test should I use?
Your recommended test
Assumptions to check first
Also consider
Get the free Statistical Test Selection toolkit
A printable decision map plus assumption-check and reporting checklists from the Statistical Test Selection Workbook. We’ll email you the download link.
How to choose a statistical test
Picking the right test comes down to four questions: What are you trying to do (compare groups, test a relationship, predict an outcome, or compare one sample to a known value)? What type is your outcome variable (numeric, ordinal, categorical, count, or time-to-event)? How many groups are you comparing? And are those groups independent or paired? The selector above turns those answers into a named test plus the assumptions worth checking.
- Numeric outcome, 2 groups, independent: independent-samples t-test (Mann–Whitney U if non-normal).
- Numeric outcome, 2 groups, paired: paired t-test (Wilcoxon signed-rank if non-normal).
- Numeric outcome, 3+ groups: one-way ANOVA (Kruskal–Wallis if non-normal).
- Categorical outcome / proportions: chi-square test (Fisher’s exact for small samples; McNemar’s for paired).
- Relationship between two variables: Pearson correlation (Spearman’s for ordinal or non-linear).
- Predicting an outcome: linear, logistic, Poisson, or Cox regression by outcome type.
Frequently asked questions
How do I choose the right statistical test?
Match the test to four things: your goal (compare groups, test a relationship, or predict an outcome), the type of your outcome variable, how many groups you compare, and whether the groups are independent or paired. The selector above walks you through those choices and names the test.
When should I use a non-parametric test?
Use a non-parametric test (Mann–Whitney U, Wilcoxon signed-rank, Kruskal–Wallis) when the outcome is ordinal, or when a numeric outcome is clearly non-normal and the sample is small. With large samples, parametric tests are fairly robust to non-normality.
Paired vs independent — which do I have?
Use a paired test when the same subjects are measured twice (before/after) or are matched. Use an independent test when the groups contain different, unrelated subjects.
Does this tool run my data or store anything?
No. It asks about the structure of your study and recommends a test — it never uploads or stores data. Everything runs in your browser.
I’m still unsure which test applies.
The recommendation is a strong starting point, not statistical advice for a specific dataset. Confirm the assumptions listed, and when a study is high-stakes, check with a statistician. The Statistical Test Selection Workbook walks through the edge cases.