Abstract

In multivariate regression analysis, it is commonly assumed that the residuals follow a multivariate normal distribution. However, this assumption is often violated in practice due to issues such as heteroscedasticity, outliers, or skewness. To address these violations, researchers frequently apply transformations to the data to improve the normality of residuals. This study utilizes the R package trafo to perform various data transformations and then assesses the normality assumption using both graphical and statistical methods. A numerical example is provided within the context of multivariate regression analysis. QQ plots are used to visually examine the distribution of residuals, while formal statistical tests-including the Shapiro-Wilk, Jarque-Bera, and Anderson-Darling tests-are employed to quantitatively evaluate deviations from normality. The results illustrate how data transformation can enhance adherence to the assumption of multivariate normality and underscore the importance of diagnostic checks in regression modeling.