Outlier detection methods may be resorted to in order to
identify discordant observations, a major shortcoming of most
methods being the underlying hypothesis of normality, or
even the requirement of knowing the underlying statistical
distribution [1,2]. Given such knowledge, the problem of
getting robust information from a reasonable number of data
may be readily solved. When a few data only are available,
difficulties are compounded by the fact that the main points of
interest are on the tails, where data quality is inevitably
poorer. Confidence or outlier identification intervals depend
on probability concerning tails, and the difficulty of working
in these regions appears evident. In fact some two centuries
elapsed, since the groundbreaking work of Abraham de
Moivre [3] on normal distribution, before a solution was
provided to some practical tail problems by William Sealy
Gosset [4] with his Student distribution.