huber.NR: Huber M-estimator iterative least squares algorithm
Description
Algorithm for calculating fully iterated or one step Huber M-estimators of location.
Usage
huber.NR(x, c = 1.28, iter = 20)
Arguments
x
A vector of quantitative data
c
Bend criterion. The value c = 1.28 gives 95 percent efficiency of the mean given normality.
iter
Maximum number of iterations
Value
Returns iterative least squares iterations which converge to Huber's emph{M}-estimator. The first element in the vector is the sample median. The second element is the Huber one-step estimate.
}
seealso{code{huber.one.step}, code{huber.mu}}
references{
Huber, P. J. (2004) emph{Robust Statistics}. Wiley.
Wilcox, R. R. (2005) emph{Introduction to Robust Estimation and Hypothesis Testing, Second Edition}. Elsevier, Burlington, MA.
}
author{Ken Aho}
examples{
x<-rnorm(100)
huber.NR(x)
}
Details
The Huber M-estimator is a robust high efficiency estimator of location that has probably been under-utilized by biologists. It is based on maximizing the likelihood of a weighting function. This is accomplished using an iterative least squares process. The Newton Raphson algorithm is used here. The function usually converges fairly quickly < 10 iterations. The function uses the Median Absolute Deviation function, mad, from MASS. Note that if MAD = 0, then NA is returned.