A non-parametric test evaluates evidence without assuming a specific underlying distribution for the data. It provides an alternative when parametric assumptions are unknown or cannot be met. section 5.2
Use non-parametric tests to make inferences about a broader population when a t-test, ANOVA, or another parametric method may produce unreliable results. section 5.0 section 5.2
Choose and communicate the method responsibly:
- Start with the question and study design rather than forcing the analysis into a familiar statistical model. section 2.1.3
- Consider non-parametric alternatives for group differences, correlations, and predictive or causal relationships. section 5.1.2
- Use the flexibility of these methods for categorical or ordinal data when the relevant parametric assumptions are not satisfied. section 5.2
- Check sample characteristics and test assumptions carefully; choosing an unsuitable method can create false group differences and costly organizational decisions. section 4.2 section 5.1.2
Before-and-after product comparison
An analyst compares daily Help Center visits and chat requests before and after a search-functionality change with an independent-samples t-test, then adds a non-parametric test for the final comparison when supported by the analysis plan. section 2.1.3
