- Non-parametric tests
- when sample size is very small (<30 for normal tests, < 10 for proportional tests, < 5 for chi-square tests)
- When the distribution is not normal
- Non-parametric Wilcoxon-signed rank test :
- Works well with paired data
- Takes the elementwise difference between 2 columns
- Take the absolute value of the differences
- Rank the absolute values
- Split them into two parts. Add all ranks for values that were less than zero and add all ranks for values that were greater than zero
- take the minimum of the two sum
- Non-parametric Wilcoxon-Mann-Whitney test:
- Works well with unpaired data of two categories
- requires to convert data from long to wide format
- Non-parametric Kruskal-Wallis test:
- Works well for data with multiple categories