Your data
Answer these and the guide tells you which test applies and what to do if normality fails.
From your descriptives output.
—
Recommended test
—
Checks
- Skewness within ±2—
- Kurtosis within ±7—
- Sample large enough for CLT—
- Shapiro-Wilk appropriate—
The mistakes this stage produces
- Relying on Shapiro-Wilk with a large sample. Above roughly N = 200 it flags trivial departures as significant. Use skewness and kurtosis instead and say so.
- Testing normality of the raw variable in regression. The assumption applies to the residuals, not the predictors.
- Transforming data to force a parametric result. Legitimate sometimes, but you must state it and justify it. Doing it silently is the problem.
- Running tests before checking anything. Output is always produced; it is not always meaningful.
Not sure this is the right calculation for your design? Send us your design and we will tell you which rule applies — at no charge.
Ask about your design