R help - Apr 2010 - Jonckheere-Terpstra test using coin package?

Is it possible to implement the Jonckheere-Terpstra test for ordered
alternatives using the coin package: Conditional Inference Procedures
in a Permutation Test Framework?

I found jonckheere.test{clinfun}, but it uses a normal approximation
when ties are present in the data.  To make this concrete, I've
include
a small dataset.  Thanks.  --Dale

Hollander and Wolfe, 1999 Table 6.6, pg 205

control <- c(40, 35, 38, 43, 44, 41)
rough <- c(38, 40, 47, 44, 40, 42)
accurate <- c(48, 40, 45, 43, 46, 44)

pieces <- list(control, rough, accurate)
n <- c(6, 6, 6)
grp <- as.ordered(factor(rep(1:length(n),n)))

library(clinfun)
jonckheere.test(unlist(pieces), grp, alternative="increasing")


Output below ...
> jonckheere.test(unlist(pieces), grp, alternative="increasing")
	Jonckheere-Terpstra test

data:
JT = 79, p-value = 0.02163
alternative hypothesis: increasing

Warning message:
In jonckheere.test(unlist(pieces), grp, alternative = "increasing") :
  TIES: p-value based on normal approximation

I am unfamiliar with how to use coin, but it looks like it should be doable -
the vignette in coin.pdf does a Page test. The wikipedia page for Page's
test explains how this is related to the Jonckheere-Terpstra test, which
seems to suggest if it can do one it should be able to do the other. Add to
that the fact that coin can do the unordered version of the J-T (which is
just Kruskal-Wallis) and again, it seems like it should be able to do it.

-- 
View this message in context:
http://r.789695.n4.nabble.com/Jonckheere-Terpstra-test-using-coin-package-tp2019893p2020340.html
Sent from the R help mailing list archive at Nabble.com.

Using coin ... answer provided by Prof. Hothorn.

control <- c(40, 35, 38, 43, 44, 41)
rough <- c(38, 40, 47, 44, 40, 42)
accurate <- c(48, 40, 45, 43, 46, 44)

pieces <- list(control, rough, accurate)
n <- c(6, 6, 6)
grp <- as.ordered(factor(rep(1:length(n),n)))

library("coin")
library("multcomp")

(y <- unlist(pieces))
k <- length(pieces)
(x <- as.ordered(factor(rep(1:k,n))))


 ### look at K. The second line just sums up.
       ff <- function(x) {
           K <- contrMat(table(x), "Tukey")[,x]
           as.vector(rep(1, nrow(K)) %*%K)
       }

independence_test(y ~ x, ytrafo = rank,
                  xtrafo = function(data) trafo(data, factor_trafo = ff),
                  alternative="greater",
                  distribution="asymptotic")

independence_test(y ~ x, ytrafo = rank,
                  xtrafo = function(data) trafo(data, factor_trafo = ff),
                  alternative="greater",
                  distribution=approximate(B=500000))

On Wed, Apr 21, 2010 at 8:23 PM, Dale Steele <dale.w.steele at gmail.com>
wrote:
> Is it possible to implement the Jonckheere-Terpstra test for ordered
> alternatives using the coin package: Conditional Inference Procedures
> in a Permutation Test Framework?
>
> I found jonckheere.test{clinfun}, but it uses a normal approximation
> when ties are present in the data. ?To make this concrete, I've
> include
> a small dataset. ?Thanks. ?--Dale
>
> Hollander and Wolfe, 1999 Table 6.6, pg 205
>
> control <- c(40, 35, 38, 43, 44, 41)
> rough <- c(38, 40, 47, 44, 40, 42)
> accurate <- c(48, 40, 45, 43, 46, 44)
>
> pieces <- list(control, rough, accurate)
> n <- c(6, 6, 6)
> grp <- as.ordered(factor(rep(1:length(n),n)))
>
> library(clinfun)
> jonckheere.test(unlist(pieces), grp, alternative="increasing")
>
>
> Output below ...
>> jonckheere.test(unlist(pieces), grp,
alternative="increasing")
>
> ? ? ? ?Jonckheere-Terpstra test
>
> data:
> JT = 79, p-value = 0.02163
> alternative hypothesis: increasing
>
> Warning message:
> In jonckheere.test(unlist(pieces), grp, alternative =
"increasing") :
> ?TIES: p-value based on normal approximation
>