medianPolish: Median polish

Description

Median polish.

Usage

"medianPolish"(X, tol=0.01, maxIter=10L, na.rm=NA, ..., .addExtra=TRUE)

Arguments

N-times-K matrix

tol

A numeric value greater than zero used as a threshold to identify when the algorithm has converged.

maxIter

Maximum number of iterations.

na.rm

If TRUE (FALSE), NAs are exclude (not exclude). If NA, it is assumed that X contains no NA values.

.addExtra

If TRUE, the name of argument X is returned and the returned structure is assigned a class. This will make the result compatible what medpolish returns.

...

Not used.

Value

overall: The fitted constant term.
row: The fitted row effect.
col: The fitted column effect.
residuals: The residuals.
converged: If TRUE, the algorithm converged, otherwise not.

Details

The implementation of this method give identical estimates as medpolish, but is about 3-5 times more efficient when there are no NA values.

Examples

Run this code


# Deaths from sport parachuting;  from ABC of EDA, p.224:
deaths <- matrix(c(14,15,14, 7,4,7, 8,2,10, 15,9,10, 0,2,0), ncol=3, byrow=TRUE)
rownames(deaths) <- c("1-24", "25-74", "75-199", "200++", "NA")
colnames(deaths) <- 1973:1975

print(deaths)

mp <- medianPolish(deaths)
mp1 <- medpolish(deaths, trace=FALSE)
print(mp)

ff <- c("overall", "row", "col", "residuals")
stopifnot(all.equal(mp[ff], mp1[ff]))

# Validate decomposition:
stopifnot(all.equal(deaths, mp$overall+outer(mp$row,mp$col,"+")+mp$resid))