Statistical test selection guide
Answer a few questions about your question, outcome and groups to get a ranked list of suitable methods, the assumptions each one needs, and a link to the Happytails.ai tool that runs it.
Runs locally in your browser; nothing is uploaded. Form answers only: question, outcome type, number of groups, pairing, the smallest group size (a whole number of at least 1), normality, equal spread and small expected counts. Downloads: a JSON report and a CSV list of candidates with their page paths. Choose the question and outcome first; group questions apply to comparisons. Up to 1,000,000 characters. Review the settings before running.
Your result
How to use Statistical test selection guide
- Choose what you want to find out and the type of outcome.
- For comparisons, give the number of groups, pairing, the smallest group size and what you know about normality and spread.
- Run the guide and read the recommended method and the alternatives.
- Open the linked tool, or download the list.
Useful next steps
How suggestions are made
Fixed rules map your answers to methods. Comparisons of continuous outcomes suggest Welch's t test or Welch's ANOVA first; small groups with doubtful normality, ordinal outcomes and counts put rank tests first. Binary and category outcomes of independent groups suggest chi-square or Fisher's exact test. Paired yes/no data get McNemar's test for two measurements and Cochran's Q, run as the Friedman test on 0 and 1 codes, for three or more; one sample of yes/no outcomes gets the exact binomial test against a reference proportion. Paired category outcomes and one sample of categories are marked not suitable, because the Stuart-Maxwell, Bowker and goodness-of-fit tests they need are not available here. Other questions lead to regression, agreement, reliability, diagnostic accuracy, distribution, time series or meta-analysis tools.
Reading the list
Each candidate is marked recommended, alternative or not suitable, with fixed sentences for assumptions you have not confirmed, such as equal variances for Student's t test. Every suggestion links to the Happytails.ai page that runs it; a method marked not suitable links back to this guide.
Limits
The rules cover the methods in this catalogue only and cannot see your data or study design. They are a starting point to check with someone who knows the study, not a statistical review.