In order to identify possible outliers, we recommend the use of robust parametric methods. In our case, for outlier detection, we use Median Absolute Deviation (MADN):

MAD = median|xi − M|i=1,...,n (M is the median of a set of n observations)

MADN = MAD/0.6745

A value xi is then assumed to be an outlier if:
|xi − M|/ MADN > k

For outlier detection, Wilcox recommends setting k to 2.24. The value 2.24 corresponds to the 0.9875 quantile of the standard normal distribution.

Reference:

Wilcox RR (2012) Introduction to robust estimation & hypothesis testing, 3rd edn. Elsevier.