What test compares two groups on a continuous outcome?

An independent-samples t-test if the data are roughly normal and the groups are separate; a paired t-test if the two measurements come from the same people; and the Mann-Whitney U (independent) or Wilcoxon signed-rank (paired) test if normality is not met.

What test do I use for more than two groups?

One-way ANOVA for independent groups on a continuous outcome (Kruskal-Wallis if assumptions fail), and repeated-measures ANOVA when the same participants are measured under each condition (Friedman test as the non-parametric alternative).

What test do I use for relationships between variables?

Pearson correlation for a linear relationship between two continuous variables (Spearman if not normal or ordinal), and regression when you want to predict an outcome from one or more predictors — linear regression for continuous outcomes, logistic regression for binary ones.

Home › Guides › Statistics › Which test?

Which statistical test should I use?

Q: How do I choose a statistical test?

Answer four questions: what type of outcome do you have (continuous, ordinal, or categorical), how many groups are you comparing, are the groups independent or paired/repeated, and are the test's assumptions (such as normality) met. Those four answers point to a specific test.

By Dr. Rafiq Muhammad, MD, PhD · Updated June 2026

The right test falls out of four facts about your study: the type of outcome you measured, how many groups you’re comparing, whether those groups are independent or paired, and whether your data meet the test’s assumptions. Answer those four and the test is decided.

Skip to the answer: the free Statistical Test Selector asks these questions and names your test (with the assumptions to check) in about 30 seconds.

The four questions

What kind of outcome? Continuous (e.g. blood pressure, score), ordinal (ranked categories), or categorical/nominal (e.g. yes/no, group membership).
How many groups or conditions? One, two, or three-plus.
Independent or paired? Different people in each group (independent) vs the same people measured more than once (paired/repeated).
Are assumptions met? Many tests assume roughly normal data (and equal variances). If not, you use a non-parametric alternative.

Decision table (the common cases)

Goal	If assumptions met	If not (non-parametric)
Compare 2 independent groups, continuous outcome	Independent-samples t-test	Mann–Whitney U
Compare 2 paired measurements, continuous	Paired t-test	Wilcoxon signed-rank
Compare 3+ independent groups, continuous	One-way ANOVA	Kruskal–Wallis
Compare 3+ repeated conditions, continuous	Repeated-measures ANOVA	Friedman
Relationship between 2 continuous variables	Pearson correlation	Spearman correlation
Association between 2 categorical variables	Chi-square test	Fisher’s exact (small samples)
Predict a continuous outcome from predictors	Linear regression	—
Predict a binary outcome from predictors	Logistic regression	—

Three traps to avoid

Treating paired data as independent (or vice versa) — the paired design is more powerful; using the wrong family throws that away.
Ignoring the outcome type — a t-test on an ordinal Likert item, or a continuous test on a yes/no outcome, is a common reviewer flag.
Skipping the assumption check — normality, equal variances, and (for chi-square) expected cell counts decide whether you need the non-parametric column.

Once you have your test, two follow-ups complete the picture: do you have enough participants to detect the effect, and how large is that effect?

Get the free Statistics toolkit

A full test-selection flowchart, assumption checklists, and worked examples from the Statistical Test Selection Workbook. We’ll email you the download link.

Get the Statistical Test Selection Workbook

Frequently asked questions

How do I choose a statistical test?

Answer four questions — outcome type, number of groups, independent vs paired, and whether assumptions are met — and the test is decided. The selector tool does this for you.

Which test for two groups, continuous outcome?

Independent t-test (or paired t-test if same people); Mann–Whitney U / Wilcoxon if not normal.

Which test for 3+ groups?

One-way ANOVA (Kruskal–Wallis if assumptions fail); repeated-measures ANOVA / Friedman for the same participants across conditions.

Which test for relationships?

Pearson (or Spearman) correlation for two continuous variables; regression to predict an outcome from predictors.

← Statistics guide Open the Test Selector →