Home/Free tools/Statistical test selector
Statistical test selector
Pubrica’s free statistical test selector recommends an appropriate test based on your outcome type, the groups compared, the study design and the data distribution. It also shows the alternative test to use when an assumption is not met, what to report and a common reviewer comment for that design.
The tool runs in your browser. No registration, email address or file upload is required, and the formula, reference or assessment criteria used are shown with the result.
- Runs in your browser
- Nothing uploaded or stored
- No email asked for
- Method shown with the answer
- Sources dated
The Tool
Select a Statistical Test for Your Study Design
Recommends a statistical test from your outcome type, groups, design and distribution, with the alternative to use when assumptions are not met.
06
Recommends a statistical test from your outcome type, groups, design and distribution, with the alternative to use when assumptions are not met.
Decision tree · 5 outcome typesIndependent-samples t-test (Welch, not Student)
Mann–Whitney U, or a t-test on transformed data
The difference between groups with a 95% confidence interval, in the units you measured — not the p-value on its own, and not the two group means with a star between them.
Welch’s version does not assume equal variances and should be the default; testing for equal variance first and then choosing inflates the error rate.
Reporting a Mann–Whitney p-value beside a difference in means — the test and the estimate then disagree about what was compared.
How to read this. A chooser gets you to a defensible starting point; it cannot see your data. Clustering, repeated measures within a cluster, a stopping rule, a non-inferiority margin, multiple primary outcomes or missing data that is not missing at random will all override the answer above — and each of them has to be handled in the analysis plan, before the data are looked at.
Limitations
What This Tool Does Not Cover
Use the result as a starting point. These are the situations in which it needs specialist review.
The calculators
Limitations of the calculators
The sample size calculator gives the closed-form answer for two independent groups. Cluster randomisation, interim analyses, a non-inferiority margin, repeated measures and competing risks each change the number materially, and none of them can be handled by inflating the answer afterwards. The same applies to the test chooser: it cannot know that your observations are clustered within clinics, and clustering overrides almost everything else it would tell you.
Related Services
Related Pubrica Services
When your project needs more than an estimate, these services provide specialist support.
More Free Tools
Other Free Research Tools
All Pubrica tools run in your browser and show the method used with each result.
Discuss Your Statistical Analysis Plan
Send us your study design, manuscript or deadline. A subject specialist will review it and provide a quote within two business days, at no cost and with no obligation.