[R] wilcox_test function in coin package
Henric Winell
nilsson.henric at gmail.com
Mon Jun 3 12:15:39 CEST 2013
Janh,
Janh Anni skrev 2013-06-01 19:47:
> Hello All,
> Thanks a lot for the helpful suggestions. I wonder how ties are handled
> for the rank sum test by wilcox_test and wilcox.exact? For instance,
Ties handling was mainly a problem back in the day when recursion
formulas were used for the computation of exact p-values. (When no ties
are present, the form of the exact null distribution of the Wilcoxon
rank-sum statistic depends only on the total number of observations in
the two groups.)
> other software such as Minitab correct for ties by adjusting the
> variance of the test statistic, and actually provide the p values before
> and after adjustment for ties.
Adjusting the variance only matters when approximating the exact
distribution with its asymptotic Gaussian distribution. And only when
at least one tie is shared between the two groups. And yes, the
standardized statistic in 'coin' accounts for ties. If ties are
present, why would you want to know the unadjusted p-value?
> IIf neither wilcox_test nor wilcox.exact
> expressly corrects for ties in the Wilcoxon rank sum test, then perhaps
> one should just use the conventional wilcox.test which is the simplest
> of them all? Thanks again
If this was true, the 'wilcox.exact' function would be completely
pointless since 'wilcox.test' just falls back on the asymptotic
approximation when ties are present.
Software like 'StatXact', 'exactRankTests', and 'coin' use algorithms
that compute the exact p-value for any ties configuration. Take a look
at Torsten Hothorn's "On Exact Rank Tests in R" article from the very
first issue of R News http://www.r-project.org/doc/Rnews/Rnews_2001-1.pdf.
Henric
> Janh
>
>
> On Sat, Jun 1, 2013 at 8:41 AM, Henric Winell <nilsson.henric at gmail.com
> <mailto:nilsson.henric at gmail.com>> wrote:
>
> Janh,
>
> Janh Anni skrev 2013-06-01 04:27:
>
> Hello peter,
>
>
> Thanks for the comment. wilcox.exact is simpler as you pointed
> out but the
> fact that it is no longer being developed is somewhat concerning.
>
>
> Admittedly, 'coin' is being actively developed and has a lot more
> bells and whistles. But for something as simple as this, that
> wouldn't bother me at all. In any case, the 'exactRankTests'
> package still gets bug fixes and the algorithm used in the Wilcoxon
> case is exactly the same for both packages.
>
> However, if you want to stay with 'coin' you can just wrap up Greg's
> proposal in a function:
>
> wilcox_test.default <- function(x, y, ...) {
> data <-
> data.frame(values = c(x, y),
> group = rep(c("x", "y"), c(length(x), length(y))))
> wilcox_test(values ~ group, data = data, ...)
> }
>
> Assuming that both 'coin' and 'exactRankTests are loaded, we can now
> check that it works:
>
> > set.seed(123)
> > x <- rpois(10, 3)
> > y <- rpois(11, 3.1)
> >
> > wilcox_test(x, y, alternative = "less", distribution = "exact")
>
> Exact Wilcoxon-Mann-Whitney Test
>
> data: values by group (x, y)
> Z = -0.0715, p-value = 0.4844
> alternative hypothesis: true mu is less than 0
>
> > wilcox.exact(x, y, alternative = "less")
>
> Exact Wilcoxon rank sum test
>
> data: x and y
> W = 54, p-value = 0.4844
> alternative hypothesis: true mu is less than 0
>
>
> HTH,
> Henric
>
>
>
>
> Regards
> Janh
>
>
> On Fri, May 31, 2013 at 6:50 PM, Peter Ehlers
> <ehlers at ucalgary.ca <mailto:ehlers at ucalgary.ca>> wrote:
>
> On 2013-05-30 20:20, Janh Anni wrote:
>
> Hello Greg,
>
> Thank you so much for your kind assistance. It looks
> like there's no way
> around using the formula format. I longed in vain for a
> simpler script
> more like the wilcox.test format. Thanks again.
>
> Janh
>
>
> I don't see why the formula syntax would be a problem, but
> to avoid it
> you could use exactRankTests::wilcox.exact() which, I
> believe, was
> written by the same author. It uses the same syntax as
> wilcox.test().
> Note, though, that the package is no longer
> being developed.
>
> Peter Ehlers
>
>
>
>
> On Thu, May 30, 2013 at 6:21 PM, Greg Snow
> <538280 at gmail.com <mailto:538280 at gmail.com>> wrote:
>
> Ok, it looks like the function mainly works through
> the formula syntax.
>
> It still would have been nice to have a
> reproducible example of what
> your
> data may look like, but I can show an example with
> simulated x and y:
>
> x <- rpois(10, 3)
>
> y <- rpois(11, 3.1)
> mydf <- data.frame( vals = c(x,y),
>
> + group=rep( c('x','y'), c( length(x), length(y) ) ) )
>
> wilcox_test( vals ~ group, data=mydf )
>
>
> Asymptotic Wilcoxon-Mann-Whitney Test
>
> data: vals by group (x, y)
> Z = -1.3718, p-value = 0.1701
> alternative hypothesis: true mu is not equal to 0
>
> Does that help? (maybe I am the heedlessness
> theorist after all)
>
>
>
> On Thu, May 30, 2013 at 4:14 PM, Janh Anni
> <annijanh at gmail.com <mailto:annijanh at gmail.com>> wrote:
>
> I thought (hoped) wilcox_test(x,y) would do it
> but it doesn't and the
>
> package maintainer says the data have to be
> rearranged but does not
> specify
> how. Thanks
>
> Janh
>
>
>
> On Thu, May 30, 2013 at 6:05 PM, Greg Snow
> <538280 at gmail.com <mailto:538280 at gmail.com>> wrote:
>
> What have you tried so far? Have you read
> the help page? have you run
>
> the examples on that page?
>
> I would expect that it is something as simple as
>
> library(coin)
> wilcox_test(x,y)
>
> or
>
> wilcox_test( y ~ group )
>
> But you should trust the help page more than
> the expectations of
> someone
> who has not read it recently (see fortune(14)).
>
> If that does not answer your question then
> give us more detail on what
> you tried, what you expected the results to
> be, what the results
> actually
> were, and how they differed. Without that
> information we have to
> resort to
> mind reading and the current implementation
> of the esp package is still
> very pre-alpha, it suggests that the answer
> to your question is:
>
> esp()
>
>
> [1] "selflessly vigilantly pigeon theorist
> heedlessness"
>
> Which is either much to profound for the
> likes of me to understand or
> is
> complete gibberish (which is only slightly
> less helpful than an overly
> general question without a reproducible
> example).
>
>
> On Thu, May 30, 2013 at 2:07 PM, Janh Anni
> <annijanh at gmail.com
> <mailto:annijanh at gmail.com>> wrote:
>
> Dear All,
>
>
> I have two simple data samples (no
> groups or factors, etc.) and would
> just
> like to compute the two-sample Wilcoxon
> Rank Sum test using the
> wilcox_test
> function contained in the coin package,
> which is reportedly better
> than
> the
> regular wilcox.test function because it
> performs some adjustment for
> ties.
> Would anyone know how to craft a script
> to perform this task? Much
> appreciated.
>
> Janh
>
> [[alternative HTML version
> deleted]]
>
> ________________________________**________________
> R-help at r-project.org
> <mailto:R-help at r-project.org> mailing list
> https://stat.ethz.ch/mailman/*__*listinfo/r-help
> <https://stat.ethz.ch/mailman/**listinfo/r-help><https://stat__ethz.ch/mailman/listinfo/r-__help
> <https://stat.ethz.ch/mailman/listinfo/r-help>>
>
> PLEASE do read the posting guide
> http://www.R-project.org/**__posting-guide.html
> <http://www.R-project.org/**posting-guide.html><http://www.__R-project.org/posting-guide.__html
> <http://www.R-project.org/posting-guide.html>>
>
> and provide commented, minimal,
> self-contained, reproducible code.
>
>
>
>
> --
> Gregory (Greg) L. Snow Ph.D.
> 538280 at gmail.com <mailto:538280 at gmail.com>
>
>
>
>
>
> --
> Gregory (Greg) L. Snow Ph.D.
> 538280 at gmail.com <mailto:538280 at gmail.com>
>
>
> [[alternative HTML version deleted]]
>
> ________________________________**________________
> R-help at r-project.org <mailto:R-help at r-project.org>
> mailing list
> https://stat.ethz.ch/mailman/*__*listinfo/r-help
> <https://stat.ethz.ch/mailman/**listinfo/r-help><https://stat__ethz.ch/mailman/listinfo/r-__help
> <https://stat.ethz.ch/mailman/listinfo/r-help>>
> PLEASE do read the posting guide http://www.R-project.org/**
> posting-guide.html
> <http://www.R-project.org/__posting-guide.html
> <http://www.R-project.org/posting-guide.html>>
>
> and provide commented, minimal, self-contained,
> reproducible code.
>
>
>
>
>
> [[alternative HTML version deleted]]
>
> ________________________________________________
> R-help at r-project.org <mailto:R-help at r-project.org> mailing list
> https://stat.ethz.ch/mailman/__listinfo/r-help
> <https://stat.ethz.ch/mailman/listinfo/r-help>
> PLEASE do read the posting guide
> http://www.R-project.org/__posting-guide.html
> <http://www.R-project.org/posting-guide.html>
> and provide commented, minimal, self-contained, reproducible code.
>
>
>
More information about the R-help
mailing list