Package 'outlying'

Grubbs' test

Description

Iteratively search for all possible outliers in a numeric vector. The default method is a modification version of Grubbs' test, which is slightly more sensitive to "far-points" compare to the original one. Please change the sensitivity to 1 if the original test is preferred.

Usage

Grubbs_test(
  x,
  alpha = 0.05,
  min_n = 7L,
  iteration = -1L,
  max_out = 0.2,
  use_median = FALSE,
  sensitivity = 2L,
  verbose = FALSE
)

Arguments

x	A numeric vector.
alpha	Default: 0.05 (two-tailed, thus 0.025 for each side).
min_n	A positive integer (default: 7). The minimum observations required for the test.
iteration	How many iterations of the test should be proceeded (default: -1; means unlimited)? Each iteration will only recognize one outlier. For example, iteration = 3 means the test will find no more than 3 outliers.
max_out	The maximum proportion (ranged from 0 to 1) of outliers to be detected in the dataset (default: 0.2, which means the data contain no more than 20% of outliers data points). If too many outliers, simply discarding them using this approach might be inappropriate.
use_median	Use the median or the mean value as the center (default: FALSE).
sensitivity	An integer value range from 1 to 3 (default: 3). The higher the value, the more sensitive of the test. Value of 1 is essentially the original Grubbs' test, which is probably too conservative. Value of 2 recalculates the mean after discarding the outlier for each iteration. Value of 3 is same with 2, but the standard deviation is also recalculated.
verbose	Should the output includes statistics result (default: FALSE)?

Value

By default (verbose = FALSE), return a logical named vector indicating the outlying elements. If verbose = TRUE, return a list which contains statistic values.

References

Grubbs, F. E. (1969). Procedures for Detecting Outlying Observations in Samples. Technometrics, 11(1), 1–21. https://doi.org/10.1080/00401706.1969.10490657

Examples

set.seed(1)
#----------------------------------------------------------------------------
Grubbs_test(c(0, 0, 7, 0, 0, 1, 0))
#>     0     0     7     0     0     1     0
#> FALSE FALSE  TRUE FALSE FALSE  TRUE FALSE
#----------------------------------------------------------------------------
x <- c(round(rnorm(3, 0, 1), 2), -5, 3)
Grubbs_test(x, min_n = 5, max_out = 0.4)
#> -0.63  0.18 -0.84    -5     3
#> FALSE FALSE FALSE  TRUE  TRUE
#----------------------------------------------------------------------------
x <- round(c(rnorm(10, 0, 1), 5))
Grubbs_test(x)
#>     2     0    -1     0     1     1     0     2     0    -1     5
#> FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE  TRUE

---