When data are non-Gaussian and parametric methods will be used, what steps must be taken?

• A. Data must be transformed and retested; If normality cannot be established after transformation, parametric methods cannot be used
• B. Data must be log-transformed only and then used with parametric methods
• C. Data should be analyzed with parametric methods regardless of normality
• D. Nonparametric methods should be used instead

Answer: (A)

When planning parametric tests, the distribution assumed is approximately normal for the data or the model residuals. If the data are non-Gaussian, you should try to bring them closer to normal with a transformation (for example, log, square root, or Box-Cox) and then re-check normality using both formal tests and visual assessments like Q-Q plots. If the transformed data (and the residual structure) meet the normality assumption, you can proceed with the parametric analysis on the transformed scale (and back-transform results for interpretation if needed). If normality still cannot be established after an appropriate transformation, parametric methods are not appropriate, and you should use nonparametric tests or other suitable modeling approaches. It’s important to note that a single log transform isn’t universally sufficient; the right transformation depends on the data, and normality must be verified after transformation before applying parametric methods.