Normality - Reasons for Screening Data
Prior to analyzing multivariate normality, one should consider univariate normality
- Histogram, Normal Q-Qplot (values on x axis with expected normal values on the y axis)
- Skewness and Kurtosis (null hypothesis: values around zero with alpha levels of .01 or .001
- Kolmogorov-Smirnov Test
Multivariate normality refers to a normal distribution of combination of variables (two-by-two, plus all linear combination of the variables) Univariate normality is a necessary but not sufficient condition for multivariate normality.
For bivariate normality one should check all the two-by-two scatter plots (they should have elliptical shape)
Sometimes data transformation is necessary for normality.