Three version of the test are implemented (univariate, angle and radius
test) and it depends on the data which test is chosen.
If the data is univariate the univariate Anderson-Darling test for normality
is applied.
If the data is bivariate the angle Anderson-Darling test for normality is
performed out.
If the data is multivariate the radius Anderson-Darling test for normality
is used.
If ‘locscatt’ is equal to “robust” then within the procedure,
robust estimates of mean and covariance are provided using ‘covMcd()’
from package robustbase.
To provide estimates for the corresponding p-values, i.e. to compute the
probability of obtaining a result at least as extreme as the one that was
actually observed under the null hypothesis, we use Monte Carlo techniques
where we check how often the statistic of the underlying data is more
extreme than statistics obtained from simulated normal distributed data with
the same (column-wise-) mean(s) and (co)variance.