Outlier detection threshold (upper) based on Tukey's rule, corrected for skewness using the 'medcouple', and possibly corrected for multiple testing.

tukey_mc_up(x, coef = NULL, alpha = 0.05, a = -4, b = 3)

Arguments

x

Numeric vector. Should be somewhat normally distributed.

coef

number determining how far 'whiskers' extend out from the box. If NULL (default), this is computed to get an type-I error of alpha, after adjusting for multiple testing. A standard value to use is 1.5.

alpha

See coef. Default is 0.05.

a, b

scaling factors multiplied by the medcouple mc() to determine outlyer boundaries; see the references.

References

Hubert, M. and Vandervieren, E. (2008). An adjusted boxplot for skewed distributions, Computational Statistics and Data Analysis 52, 5186--5201. doi:10.1016/j.csda.2007.11.008

Examples

hist(x <- c(rnorm(3, m = 6), rnorm(1e4, m = 0)))
(q <- tukey_mc_up(x))
#> [1] 4.200123
abline(v = q, col = "red")

which(x > q)
#> [1] 1 2 3