Statistics made easy: how to choose the right statistical test

Three questions before any test

Before opening SPSS, STATA, R, or Jamovi, answer three questions: what type of outcome are you measuring (continuous, categorical, time-to-event), how many groups or predictors are involved, and is your data independent or repeated on the same participants?

A simple decision path

1. Comparing two groups on a continuous outcome — t-test (or Mann-Whitney U if not normally distributed).
2. Comparing more than two groups — ANOVA (or Kruskal-Wallis for non-parametric data).
3. Testing a relationship between two continuous variables — correlation, then linear regression if you want to predict or adjust for other factors.
4. A binary outcome (yes/no) — logistic regression, especially once you need to adjust for confounders.
5. Time-to-event data — survival analysis (Kaplan-Meier curves, Cox regression).

Common pitfalls

• Running many tests without a pre-specified analysis plan increases the risk of false positives.
• Ignoring assumptions (normality, independence) can invalidate results even when the software runs without error.
• Reporting p-values without effect sizes or confidence intervals limits how useful your results are to readers.

A short, pre-specified statistical analysis plan—written before data collection—resolves most of these issues in advance.