R/tbrm.R

In bhklab/genefu: Computation of Gene Expression-Based Signatures in Breast Cancer

#' @title Function to compute Tukey's Biweight Robust Mean
#'
#' @description
#' Computation of Tukey's Biweight Robust Mean, a robust average that is 
#'   unaffected by outliers.
#'
#' @usage
#' tbrm(x, C = 9)
#'
#' @param x	a numeric vector
#' @param C	a constant. C is preassigned a value of 9 according to the Cook 
#'   reference below but other values are possible.
#'
#' @details
#' This is a one step computation that follows the Affy whitepaper below see 
#'   page 22. This function is called by chron to calculate a robust mean. C 
#'   determines the point at which outliers are given a weight of 0 and 
#'   therefore do not contribute to the calculation of the mean. C=9 sets 
#'   values roughly +/-6 standard deviations to 0. C=6 is also used in 
#'   tree-ring chronology development. Cook and Kairiukstis (1990) have 
#'   further details.
#' Retrieved from tbrm.
#'
#' @return
#' A numeric mean.
#' 
#' @references
#' Statistical Algorithms Description Document, 2002, Affymetrix. p22.
#' Cook, E. R. and Kairiukstis, L.A. (1990) Methods of Dendrochronology: 
#'   Applications in the Environmental Sciences. ISBN-13: 978-0792305866.
#' Mosteller, F. and Tukey, J. W. (1977) Data Analysis and Regression: 
#'   a second course in statistics. Addison-Wesley. ISBN-13: 978-0201048544.
#'
#' @seealso
#' chron
#'
#' @examples
#' tbrm(rnorm(100))
#'
#' @md
#' @export
tbrm <-
function(x,C=9)
{
    x=x[!is.na(x)]
    wt=rep(0, length(x))
    x.med=median(x)
    S.star=median(abs(x - x.med))
    w0=(x - x.med)/(C * S.star + 1e-06)
    lt0.flag=abs(w0) <= 1
    wt[lt0.flag]=((1 - w0^2)^2)[lt0.flag]
    t.bi.m=sum(wt * x)/sum(wt)
    t.bi.m
}