[Rd] KS test in R.1.3.0 has incorrect p-values. (PR#1004)

[Kurt Hornik]

>>>>> ripley  writes:

> Based on a report to the Windows maintainers from Richard Rowe
> <Richard.Rowe@jcu.edu.au>:

> NEWS for 1.3.0 says

>     o   Exact p-values are available for the two-sided two-sample
>         Kolmogorov-Smirnov test.

> I think the (new) p-values are computed but are backwards:

>> set.seed(123)
>> x <- rnorm(50)
>> y <- runif(50)
>> ks.test(x,y, exact=T)$p
> [1] 1
>> 1 - ks.test(x,y, exact=T)$p
> [1] 4.047605e-08
>> ks.test(x,y, exact=F)$p
> [1] 3.099506e-07
>> y<-rnorm(50)
>> ks.test(x,y, exact=T)$p
> [1] 0.02015456
>> ks.test(x,y, exact=F)$p
> [1] 0.9971923

> I'll commit the obvious fix unless anyone knows better.

Thanks for doing.  I had received a similar private bug report but had
not gotten to fixing this.

> Neither value agrees well with S-PLUS 6, which gives

>> source("dumpdata.R")
>> ks.gof(x, y)$p
> [1] 1.453458e-07

Strange because the underlying algorithm should be the same.  Have you
tried R vs Splus on larger samples (say, 100 each)?

-k
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._