How to choose a statistical test without guessing from a chart
The correct test starts with the claim you want to evaluate and the data you actually collected. A decision tree can help, but it cannot replace checking the design and assumptions.
Name the analytical goal
Decide whether the question asks about a difference, association, prediction, classification, change over time, or another goal. Then identify the outcome variable and the variable or groups used to explain it.
Write the intended result in plain language before naming a test. For example: compare the average score of two independent groups, or estimate the association between two continuous measures.
Identify the variable types and design
Record whether each variable is categorical, ordinal, count, or continuous. Then check whether observations are independent, paired, repeated, clustered, or time ordered. The same variable types can require different models under different designs.
- Outcome variable
- Predictor or grouping variable
- Number of groups
- Independent or paired data
- Repeated measures
- Sample size
Check assumptions before interpreting
Assumptions concern the model and design, not a ritual list of tests. Depending on the analysis, check independence, distribution of residuals, variance, linearity, expected counts, influential observations, and missing-data patterns.
If an assumption is not met, consider whether transformation, a robust method, a nonparametric alternative, or a different model is justified. Report the decision and its reason.
Report an answer, not only a p-value
Connect the result to the research question. Report the estimated effect, uncertainty such as a confidence interval, the test or model used, relevant diagnostics, and practical meaning. Statistical significance alone does not show that an effect is large or useful.
Questions students ask next.
Should I choose a test before collecting data?+
Yes, at least provisionally. Planning the analysis helps ensure the design and sample can answer the question. Confirm the final method after checking the collected data.
Is a nonparametric test always safer?+
No. It answers a particular statistical question and has its own assumptions. Choose it because it fits the design and estimand, not as an automatic fallback.