Silent incorrect results for data with missing values (NAs) ?

[karoliskoncevicius]

Seems like wilcoxauc() silently returns incorrect results when data has missing values:

library(presto)
set.seed(14)

x <- rnorm(20)
g <- gl(2, 10)

All is good when no missing values are present:

wilcox.test(x ~ g, exact=FALSE)$p.value
[1] 0.7913368
wilcoxauc(t(x), g)$pval[1]
[1] 0.7913368

Add one missing obvervation:

x[14] <- NA

Now the result is not correct:

wilcox.test(x ~ g, exact=FALSE)$p.value
[1] 0.9674355
wilcoxauc(t(x), g)$pval[1]
[1] 0.9097219

Remove the missing value from the data:

x <- x[-14]
g <- g[-14]

Result is correct again:

wilcox.test(x ~ g, exact=FALSE)$p.value
[1] 0.9674355
wilcoxauc(t(x), g)$pval[1]
[1] 0.9674355