[R] problematic p values for Dixon Q test

[ronny]

I am using R dixon.test in order to perform Q test for detection of outliers.
The p-values for high Q values are consistent with the table in Rorabacher,
D.B. (1991) http://www.flworkshop.com/sscs/dixon_test.pdf, but it seems that
small Q-values get problematic p-values. For example:

 dixon.test(c(0.324,0.5,1,1.324),two.sided=T)
Q = 0.324, p-value = 1

 dixon.test(c(0.324,0.5,1,1.324),two.sided=F)
Q = 0.324, p-value = 0.5

The p value I expect by simulation is ~0.83:

 norm=matrix(nrow=1000,ncol=4)
 for (i in 1:1000){  
 norm[i,]<-rnorm(4)  
 q[i]<-dixon.test(norm[i,])$statistic 
 p[i]<-dixon.test(norm[i,])$p.value>}<br>
 hist(q) 
 expected.p.val<- length(which(q>0.324))/1000

In particular I am concerened from the misleading result of
 dixon.test(c(1,1,2,2),two.sided=T)
 Q = 0, p-value < 2.2e-16

I would appreciate any help with this issue,




--
View this message in context: http://r.789695.n4.nabble.com/problematic-p-values-for-Dixon-Q-test-tp4642983.html
Sent from the R help mailing list archive at Nabble.com.