NotebookTest-selection table
Test-selection table
Appendix 1 of the mock paper: which statistical test to use for your outcome, groups and pairing.
Choose your outcome, the number of groups and whether the data are paired. The matching row and column light up; the words are exactly as printed. For how to make each choice, see Choosing a statistical test.
Test to be used
Choose all three to see the test.
| Type of outcome variable | Normality | Paired1 | Test to be used (two groups) | Test to be used (three or more groups) |
|---|---|---|---|---|
| Nominal | - | No | Fisher's exact test or Chi-square test2 | Chi-square test |
| Nominal | - | Yes | McNemar test (binary responses) and regression modelling for multiple responses | Cochran Q-test (for binary responses) and regression modelling for multiple responses |
| Ordinal (few categories) | - | No | Chi-square test for trend/Cochran-Armitage Trend Test (binary responses) or Cochran-Mantel-Haenszel test (multiple responses) | Regression modelling |
| Ordinal (few categories) | - | Yes | Cochran-Mantel-Haenszel test | Regression modelling |
| Ordinal (multiple categories), counted variables (discrete), continuous (non-normal) | No | No | Mann-Whitney U test (also called Wilcoxon rank-sum test) | Kruskal-Wallis test |
| Ordinal (multiple categories), counted variables (discrete), continuous (non-normal) | No | Yes | Wilcoxon signed-rank test (this is different from Wilcoxon rank-sum test) | Friedman test |
| Continuous (normal) | Yes | No | Unpaired T-test (Welch correction if different SD assumed) | ANOVA |
| Continuous (normal) | Yes | Yes | Paired sample T-test | Repeated measures ANOVA |
| Time-to-event | No | No | Log-rank test | Log-rank test |
- We recommend regression modelling for paired data to compare between groups.
- Caution: when expected frequency for an event or no event <5, total sample size < 50, the results may be unreliable
Mock Appendix 1 · L6 p.51 · PBL2 answers
About this copy
- Footnote 1 is attached to the 'Paired' column header; footnote 2 to 'Chi-square test' in the first row.
- '-' in the Normality column is printed on the paper and not explained (presumably not applicable).
- Footnote 2 does not say whether both conditions or either one triggers the caution.
- The table lists the Cochran-Mantel-Haenszel test for paired ordinal (few categories) two-group data; reproduced as printed.