medpolish: Median Polish (Robust Twoway Decomposition) of a Matrix

Description

Fits an additive model (twoway decomposition) using Tukey's median polish procedure.

Usage

medpolish(x, eps = 0.01, maxiter = 10, trace.iter = TRUE,
          na.rm = FALSE)

Arguments

a numeric matrix.

eps

real number greater than 0. A tolerance for convergence: see ‘Details’.

maxiter

the maximum number of iterations

trace.iter

logical. Should progress in convergence be reported?

na.rm

logical. Should missing values be removed?

Value

An object of class medpolish with the following named components:

overall

the fitted constant term.

row

the fitted row effects.

col

the fitted column effects.

residuals

the residuals.

name

the name of the dataset.

Details

The model fitted is additive (constant + rows + columns). The algorithm works by alternately removing the row and column medians, and continues until the proportional reduction in the sum of absolute residuals is less than eps or until there have been maxiter iterations. The sum of absolute residuals is printed at each iteration of the fitting process, if trace.iter is TRUE. If na.rm is FALSE the presence of any NA value in x will cause an error, otherwise NA values are ignored.

medpolish returns an object of class medpolish (see below). There are printing and plotting methods for this class, which are invoked via by the generics print and plot.

References

Tukey, J. W. (1977). Exploratory Data Analysis, Reading Massachusetts: Addison-Wesley.

Examples

Run this code

# NOT RUN {
require(graphics)

## Deaths from sport parachuting;  from ABC of EDA, p.224:
deaths <-
    rbind(c(14,15,14),
          c( 7, 4, 7),
          c( 8, 2,10),
          c(15, 9,10),
          c( 0, 2, 0))
dimnames(deaths) <- list(c("1-24", "25-74", "75-199", "200++", "NA"),
                         paste(1973:1975))
deaths
(med.d <- medpolish(deaths))
plot(med.d)
## Check decomposition:
all(deaths ==
    med.d$overall + outer(med.d$row,med.d$col, "+") + med.d$residuals)
# }

Run the code above in your browser using DataLab

Last chance! 50% off unlimited learning