Picking the wrong statistical test is one of the most common mistakes in thesis and research writing. Here's a quick decision framework:
1) What's your outcome variable? Continuous → parametric family (t-test/ANOVA) if roughly normal, otherwise rank-based (Mann-Whitney/Kruskal-Wallis). Categorical → Chi-square or Fisher's exact (use Fisher's when any expected cell count < 5).
2) How many groups? 2 groups → t-test (independent or paired) or Mann-Whitney. 3+ groups → ANOVA or Kruskal-Wallis, followed by a post-hoc test if the omnibus result is significant.
3) Repeated measures on the same subjects? Use paired/repeated-measures variants (paired t-test, repeated-measures ANOVA, Friedman) instead of independent-sample versions — treating repeated measures as independent samples is a common validity error reviewers flag.
4) Report effect size, not just p-values. APA style expects Cohen's d, eta-squared, or r alongside the test statistic — a p-value alone doesn't tell a reader how large the effect is.
If you want to run these without setting up R/SPSS, statmate.org has free versions of all of the above (calculator-only, no signup for the calculation itself).
Top comments (0)