Statistics
Which statistical test should I use?
A practical map from the shape of your experiment to the test, with lab examples.
| Your question | Data meet assumptions | Skewed, ordinal or small n | Example |
|---|---|---|---|
| Two independent groups | Unpaired t-test (Welch if SDs differ) | Mann-Whitney | ALT in control vs treated mice |
| Same subjects, two conditions | Paired t-test | Wilcoxon signed-rank | Glucose before and after treatment |
| Three or more independent groups | One-way ANOVA + Tukey or Dunnett | Kruskal-Wallis + Dunn | Control, disease, disease + drug |
| Two factors | Two-way ANOVA | (transform, or analyse factors separately) | Diet × drug |
| Measured repeatedly over time | Repeated-measures ANOVA (Greenhouse-Geisser) or a mixed model | Friedman (one group) | Weekly body weight, GTT curve |
| Proportions / counts | Chi-square | Fisher’s exact (small counts) | Responders by arm |
| Time to an event | Kaplan-Meier with log-rank; Cox regression for hazard ratios | Survival after sepsis | |
| Diagnostic accuracy | ROC curve, AUC with 95% CI | FIB-4 for advanced fibrosis | |
| Agreement between methods | Bland-Altman (continuous); Cohen’s kappa (categories) | Point-of-care vs lab glucose | |
Three questions that settle most cases
- How many groups, and are they independent? The same animals measured twice are paired, not independent.
- What kind of outcome? Continuous, ordinal (scores), categorical, or time-to-event.
- Are assumptions reasonable? Roughly normal residuals and similar spread. With very small groups, assumption tests have little power, so prior knowledge of the variable matters.
Common mistakes
- Running many t-tests instead of ANOVA with a correction for multiple comparisons.
- Treating technical replicates (wells, sections) as independent animals.
- Analysing histology scores as continuous data with a t-test.
- Reporting only p-values: give means or medians, spread, n and the exact test.
Published 2026 by Dr. Ambuj Shahi. This guide is general information for researchers; follow your institution’s and journal’s requirements.