DescTools (version 0.99.39)

Winsorize: Winsorize (Replace Extreme Values by Less Extreme Ones)

Description

Winsorizing a vector means that a predefined quantum of the smallest and/or the largest values are replaced by less extreme values. Thereby the substitute values are the most extreme retained values.

Usage

Winsorize(x, minval = NULL, maxval = NULL, probs = c(0.05, 0.95),
          na.rm = FALSE, type = 7)

Arguments

x

a numeric vector to be winsorized.

minval

the low border, all values being lower than this will be replaced by this value. The default is set to the 5%-quantile of x.

maxval

the high border, all values being larger than this will be replaced by this value. The default is set to the 95%-quantile of x.

probs

numeric vector of probabilities with values in [0,1] as used in quantile.

na.rm

should NAs be omitted to calculate the quantiles? Note that NAs in x are preserved and left unchanged anyway.

type

an integer between 1 and 9 selecting one of the nine quantile algorithms detailed in quantile to be used.

Value

A vector of the same length as the original data x containing the winsorized data.

Details

The winsorized vector is obtained by

$$g(x) = \left\{\begin{array}{ll} -c &\textup{for }x \le -c\\ x &\textup{for } |x| < c\\ c &\textup{for }x \ge c \end{array}\right. $$

You may also want to consider standardizing (possibly robustly) the data before you perform a winsorization.

See Also

Winsorize from the package robustHD contains an option to winsorize multivariate data

scale, RobScale

Examples

# NOT RUN {
## generate data
set.seed(1234)     # for reproducibility
x <- rnorm(10)     # standard normal
x[1] <- x[1] * 10  # introduce outlier

## Winsorize data
x
Winsorize(x)

# use Large and Small, if a fix number of values should be winsorized (here k=3):
Winsorize(x, minval=tail(Small(x, k=3), 1), maxval=head(Large(x, k=3), 1))
# }