Answer a few questions about your study and data, and get the right statistical test — with its assumptions, non-parametric alternative, software commands, and an example write-up.
Tip: not sure whether your data are normally distributed? Choose “Not sure” and the selector shows you both the parametric test and its non-parametric alternative, so you’re covered either way.
The Statistical Test Selector is a free, browser-based decision tool that tells you which statistical test to use for your research question. Instead of scrolling through textbooks or guessing, you answer a short sequence of questions — your analysis goal, the type of outcome variable, how many groups you’re comparing, whether they’re independent or paired, and whether your data are normally distributed — and it returns the correct test along with everything you need to run and report it.
It’s built for researchers, graduate students, dissertation and thesis writers, and anyone preparing a manuscript for publication who needs to choose the right statistical test quickly and defend that choice to a supervisor, examiner, or peer reviewer. Every recommendation comes with the test’s key assumptions, its non-parametric alternative, the exact SPSS, R, and Python commands, and a fill-in-the-blank example results write-up. Nothing you enter is uploaded or stored — the entire decision tree runs locally in your browser.
Choosing a statistical test comes down to the shape of your data and what you’re trying to find out. Work through these five questions — they’re the same ones the selector asks — and you’ll land on the correct test almost every time.
Are you comparing groups or conditions (e.g. treatment vs control), measuring a relationship between two variables (correlation), predicting or modelling an outcome from one or more predictors (regression), or assessing agreement or reliability (raters, repeated measures, scale consistency)? Your goal narrows the whole family of possible tests.
The measurement level of your outcome is the single biggest factor. Is it continuous (blood pressure, test score, reaction time), ordinal (a Likert 1–5 rating, a stage or grade), nominal / categorical (yes/no, blood type), a count (number of admissions), or time-to-event (survival data, possibly censored)? A continuous outcome points toward t-tests, ANOVA, or regression; categorical outcomes point toward chi-square or logistic regression.
Count your groups or conditions — one, two, or three or more — and decide whether they’re independent (different participants in each group) or paired / repeated (the same participants measured more than once, such as pre vs post). Two independent groups call for an independent-samples t-test; two paired measurements call for a paired-samples t-test; three or more groups move you to ANOVA or its rank-based equivalent.
This decides parametric vs non-parametric. Check normality with a histogram, a Q–Q plot, or a Shapiro–Wilk test. If a continuous outcome is roughly normal, use the parametric test (t-test, ANOVA, Pearson). If it’s skewed, has outliers, or is ordinal, use the non-parametric alternative (Mann–Whitney U, Kruskal–Wallis, Spearman). Not sure yet? The selector shows both so you can decide after checking.
Count outcomes use Poisson or negative binomial regression; time-to-event data use Kaplan–Meier and the log-rank test or Cox regression; rater and scale reliability use Cohen’s or Fleiss’ kappa, the intraclass correlation coefficient (ICC), or Cronbach’s alpha. The selector routes these automatically once you pick the matching goal.
Use this table as a fast reference for the most common research scenarios. It maps your goal, data type, and design to the recommended statistical test and its non-parametric alternative. Bookmark it — or better, run your exact case through the selector above to get assumptions and software commands too.
| Your goal | Outcome / data type | Design | Recommended test | Non-parametric alternative |
|---|---|---|---|---|
| Compare two groups | Continuous | Independent | Independent-samples t-test | Mann–Whitney U test |
| Compare two conditions | Continuous | Paired / repeated | Paired-samples t-test | Wilcoxon signed-rank test |
| Compare 3+ groups | Continuous | Independent | One-way ANOVA | Kruskal–Wallis H test |
| Compare 3+ conditions | Continuous | Repeated | Repeated-measures ANOVA | Friedman test |
| Compare to a known value | Continuous | One sample | One-sample t-test | Wilcoxon signed-rank (one sample) |
| Compare two groups | Categorical (nominal) | Independent | Chi-square test of independence | Fisher’s exact test (small counts) |
| Compare two conditions | Binary | Paired | McNemar test | — |
| Measure a relationship | Two continuous | — | Pearson correlation | Spearman’s rank correlation |
| Measure a relationship | Two ordinal | — | Spearman’s rank correlation | Kendall’s tau |
| Measure a relationship | Two categorical | — | Chi-square test of independence | Cramér’s V (effect size) |
| Predict / model | Continuous outcome | — | Linear regression | — |
| Predict / model | Binary outcome | — | Binary logistic regression | — |
| Predict / model | Count outcome | — | Poisson regression | Negative binomial (if overdispersed) |
| Compare survival | Time-to-event | Two or more groups | Log-rank test (Kaplan–Meier) | Cox proportional-hazards regression |
| Assess agreement | Two raters, categorical | — | Cohen’s kappa | Weighted kappa (ordinal) |
| Assess reliability | Multi-item scale | — | Cronbach’s alpha | McDonald’s omega |
Always confirm each test’s assumptions on your own data before running it. This table is a starting point, not a substitute for checking normality, variance, and cell counts.
This is the choice that trips up most students. Parametric tests (t-tests, ANOVA, Pearson correlation) assume your continuous data are approximately normally distributed and meet a few other conditions; when those hold, they’re more powerful. Non-parametric tests (Mann–Whitney U, Kruskal–Wallis, Spearman) work on ranks, make fewer assumptions, and are the safer choice for skewed data, ordinal scales, small samples, or data with outliers.
| Question | Parametric test | Non-parametric alternative |
|---|---|---|
| Two independent groups | Independent t-test | Mann–Whitney U test |
| Two paired measurements | Paired t-test | Wilcoxon signed-rank test |
| Three or more groups | One-way ANOVA | Kruskal–Wallis H test |
| Repeated measurements (3+) | Repeated-measures ANOVA | Friedman test |
| Correlation between variables | Pearson’s r | Spearman’s rho / Kendall’s tau |
Rule of thumb: if your outcome is continuous and passes a normality check, start parametric. If it’s ordinal, clearly skewed, or your sample is small, use the non-parametric alternative. When in doubt, run both — if they agree, report the parametric result; if they disagree, the non-parametric result is usually the more defensible one.
Choose and justify the analysis for your methods chapter — and be ready when your committee asks why you picked that test.
Match your test to your design before submission so reviewers can’t reject you on statistical grounds.
Handle group comparisons, categorical outcomes, survival data, and inter-rater reliability with the correct test each time.
Get the right approach for Likert-scale, ordinal, and repeated-measures data that rarely meet normality.
Learn the logic of test selection with plain-language explanations and assumption checklists.
Re-check a reviewer’s “wrong test” comment and switch to the defensible alternative fast.
Match the test to four things: your goal (compare, relate, predict, or assess agreement), the type of your outcome variable (continuous, ordinal, nominal, count, or time-to-event), the number of groups and whether they’re independent or paired, and whether your data are normally distributed. The selector above walks you through exactly these questions and returns the correct test.
Parametric tests (t-test, ANOVA, Pearson correlation) assume your continuous data are approximately normally distributed and are more powerful when that holds. Non-parametric tests (Mann–Whitney U, Kruskal–Wallis, Spearman) work on ranks, make fewer assumptions, and are the right choice for skewed data, ordinal scales, small samples, or data with outliers.
For two independent groups with a continuous, normally distributed outcome, use an independent-samples t-test (non-parametric alternative: Mann–Whitney U). For two paired measurements — the same participants measured twice — use a paired-samples t-test (non-parametric alternative: Wilcoxon signed-rank). For a categorical outcome, use a chi-square test or Fisher’s exact test.
Likert responses are ordinal, so non-parametric methods are usually the safest choice: Mann–Whitney U or Kruskal–Wallis to compare groups, and Spearman’s rho or Kendall’s tau for correlations. Some researchers treat multi-item Likert scales as continuous when the scale is well-validated, but single items should be analysed as ordinal.
Yes. Every recommended test lists the assumptions you must verify — such as normality, independence of observations, equal variances, or minimum expected cell counts. Check them on your own data (histograms, Q–Q plots, Levene’s test, Shapiro–Wilk) before you trust and report the result.
Yes on both counts. The tool is completely free with no sign-up, and the entire decision tree runs locally in your browser. You only select descriptions of your data — no dataset is uploaded — and nothing you choose is stored or sent anywhere.
No — it identifies the right test and gives you the SPSS, R, and Python commands to run it yourself. If you’d like an expert to run the analysis, verify assumptions, and write up the results, our statistical analysis service can do that for you.
Each recommendation includes the menu path for SPSS and the equivalent one-line command for R and Python (SciPy and statsmodels). The test itself is the same across software — you just use whichever package you have.
The selector points you to the right test — but if you’d rather have an expert run the analysis, verify your assumptions, or write up the results to journal standard, our team does exactly that. We handle everything from choosing the test to interpreting the output and drafting a publication-ready results section.
Tell us a bit about your project and we’ll get back to you within 24 hours.