depth (version 2.1-1.1)

trmean: Classical-like depth-based trimmed mean

Description

Computes a sample trimmed mean based on the Tukey depth, the Liu depth or the Oja depth.

Usage

trmean(x, alpha, W = function(dep, alpha){return(1)}, 
   method = "Tukey", ndir = 1000, approx = FALSE, 
   eps = 1e-8, ...)

Arguments

x

The data as a matrix, data frame or list. If it is a matrix or data frame, then each row is viewed as one bivariate observation. If it is a list, all components must be numerical vectors of equal length (coordinates of observations).

alpha

Outer trimming fraction (0 to 0.5). Observations whose depth is less than alpha to be trimmed.

W

Nonnegative weight function defined on [0, 1] through its argument dep. Number of arguments can be greater than 2 but the trimming fraction has to be one argument. See examples.

method

Character string which determines the depth function used. method can be "Tukey" (the default), "Liu" or "Oja".

ndir

Positive integer. Number of random directions used when approximate Tukey depth is utilised. Used jointly with approx = TRUE.

approx

Logical. If dimension is 3, should approximate Tukey depth be used? Useful when sample size is large.

eps

Error tolerance to control the calculation.

...

Any additional arguments to the weight function.

Value

Multivariate depth-based trimmed mean

Details

Dimension 2 or higher when method is "Tukey" or "Oja"; dimension 2 only when method is "Liu". Exactness of calculation depends on method. See depth.

References

Masse, J.C and Plante, J.F. (2003), A Monte Carlo study of the accuracy and robustness of ten bivariate location estimators, Comput. Statist. Data Anal., 42, 1--26.

Masse, J.C. (2008), Multivariate Trimmed means based on the Tukey depth, J. Statist. Plann. Inference, in press.

Rousseeuw, P.J. and Ruts, I. (1996), Algorithm AS 307: Bivariate location depth, Appl. Stat.-J. Roy. St. C, 45, 516--526.

See Also

med for medians and ctrmean for a centroid trimmed mean.

Examples

# NOT RUN {
## exact trimmed mean with default constant weight function
data(starsCYG, package = "robustbase")
trmean(starsCYG, .1)

## another example with default constant weight function
set.seed(159); library(MASS)
mu1 <- c(0,0); mu2 <- c(6,0); sigma <- matrix(c(1,0,0,1), nc = 2)
mixbivnorm <- rbind(mvrnorm(80, mu1, sigma), mvrnorm(20, mu2, sigma))
trmean(mixbivnorm, 0.3)

## trimmed mean with a non constant weight function
W1 <-function(x,alpha,epsilon) {
  (2*(x-alpha)^2/epsilon^2)*(alpha<=x)*(x<alpha+epsilon/2)+
  (-2*(x-alpha)^2/epsilon^2+4*(x-alpha)/epsilon-1)*
  (alpha+epsilon/2<=x)*(x<alpha+epsilon)+(alpha+epsilon<=x)
}
set.seed(345)
x <- matrix(rnorm(210), nc = 3)
trmean(x, .1, W = W1, epsilon = .05)

## two other examples of weighted trimmed mean
set.seed(345)
x <- matrix(rnorm(210), nc = 3)
W2 <- function(x, alpha) {x^(.25)}
trmean(x, .1, W = W2)
W3 <- function(x, alpha, beta){1-sqrt(x)+x^2/beta}
trmean(x, .1, W = W3, beta = 1)
# }