One hypothesis at a time
Pick one hypothesis and name its outcome (what you measured) and its predictor (what you manipulated or compared). Run through this once per hypothesis.
At a glance
| Outcome | Predictor and design | Test | Effect size | If assumptions fail |
|---|---|---|---|---|
| Continuous | 2 groups, different people | Welch’s t | d [CI] | Mann–Whitney |
| Continuous | 2 conditions, same people | Paired t | mean difference [CI], dz | Wilcoxon signed-rank |
| Continuous | 3+ groups, different people | One-way ANOVA, then Tukey or planned contrasts | η² | Kruskal–Wallis |
| Continuous | 2+ factors, different people | Two-way (factorial) ANOVA | partial η² | Regression with robust SEs |
| Continuous | Same people in several conditions, or nested data | Mixed-effects model | β [CI] | Friedman (one factor) |
| Continuous | A continuous predictor, or several predictors | Linear regression | β [CI], R² | Spearman (one predictor) |
| Two categories | Groups | χ² (Fisher’s exact for small counts) | φ | Fisher’s exact |
| Two categories | Continuous or several predictors | Logistic regression | odds ratio [CI] | — |
| 3+ categories | Groups | χ² on the full table | Cramér’s V | Fisher’s exact |
| One ordinal item | Groups | Mann–Whitney (2) or Kruskal–Wallis (3+) | rank-biserial r | — |
Notes
- A starting point. The picker matches the test to your variables and design. Your hypothesis still decides which comparison matters, so name it before you look at the data.
- Unit of analysis first. If each person, story, or item contributes several rows, either average up to that unit or use a mixed-effects model. Treating those rows as independent inflates your sample.
- Report the size. Every path ends with an effect size and its CI. Write the estimate before the p-value, as on Oct 30.
- Run it. Each result links to an example in the R lab on practice set A or B. Change the variable names and it runs on your own CSV.