Independence means that one data point does not affect the probability of another. Standard parametric tests generally assume independent data points unless the test explicitly accommodates repeated measurements, as a repeated-measures t-test or ANOVA does. section 4.2.1
Participant interactions can put independence in doubt:
- Survey respondents change their answers after hearing their peers’ answers. section 4.2.1
- Randomized A/B test participants at the same company compare the interfaces they received. section 4.2.1
- Control-group participants notice better outcomes among intervention participants. section 4.2.1
Before choosing a standard parametric test, ask whether observations could influence one another. Random assignment alone does not rule out that possibility. section 4.2.1
