Skip to content

Choosing a statistical test

Verifiedintermediate8 min

A decision path from your design to the right test, and what to do when the assumptions fail.

Start from your design, not from the test you have heard of.

Two independent groups, numeric outcome → independent-samples t-test. Assumptions violated, or n small → Mann–Whitney U.

Same units measured twice → paired t-test. Non-normal differences → Wilcoxon signed-rank.

Three or more independent groups → one-way ANOVA, followed by a post-hoc test only if the ANOVA is significant. Non-normal → Kruskal–Wallis.

Two categorical variables → chi-square test of independence. Any expected count below 5 → Fisher's exact test.

Two numeric variables → Pearson correlation if the relationship looks linear and the data are roughly normal; Spearman otherwise.

Three rules that matter more than the choice itself:

  1. Decide the test before you see the data.
  2. Always report the effect size and a confidence interval, not just p. "Significant" without a magnitude tells the reader nothing about whether it matters.
  3. n = 3 per group almost never supports a strong claim. Say so honestly rather than hoping nobody notices.

More in Statistics

Verifiedbeginner

StatQuest

Short, genuinely clear video explanations of statistical concepts, from the basics through to machine-learning methods.

  • video
  • explainers
  • statistics