[R] ks.test and wilcoxon.test results differ from other stat.packages
ruipbarradas at sapo.pt
Fri Feb 1 16:33:31 CET 2013
As for the KS test, the op might also want to look at
Simard, L'Ecuyer (2011)
for an account on the different algorithms available, and their accuracy.
Apparently, R uses different algorithms according to the test in
question. See the details section of ?ks.test. See also the references
there, such as
Marsaglia, Tsang, Wang (2003)
Hope this helps,
Em 01-02-2013 11:24, Meyners, Michael escreveu:
> Impossible to say w/o a reproducible example, but to start with let me suggest looking at the exact= (both functions) and correct= (wilcox.test) arguments. Experience shows that some change of the default settings allows you to reproduce results from other software (and the help pages will explain you what the settings do; by that, you might also learn what your other tools actually calculate if you have no documentation at hand).
> HTH, Michael
>> -----Original Message-----
>> From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-
>> project.org] On Behalf Of Alexander Favorov
>> Sent: Freitag, 1. Februar 2013 09:00
>> To: r-help at r-project.org
>> Subject: [R] ks.test and wilcoxon.test results differ from other
>> Probably, it's an obvious info, but I have not found anything in R FAQ
>> concerning this feature/bug.
>> The results of ks.test and wilcoxon.test (in the Mann-Whitney version,
>> paired = 'FALSE') don't coincide with the results from the other
>> statistical packages, e.g. Statistica, Medcalc, and (as for MW test)
>> from the numerous online MW tests.
>> Statistica p-value=0.0435353
>> Medcalc p-value=0.0435354
>> R p-value=0.04635
>> If I want to obtain result of test once, it does not matter.
>> But what should I do if I want to perform Monte-Carlo simulations and I
>> need in 10000 or even 100 000 p-values and then will build some
>> distribution and then use results of Statictica? Whtever, the
>> descrepancy bothers.
>> Are there some alternative packages for non-parametric statistics that
>> produce results comparable with the other program packages? Or,
>> probably, there is exists some environment variable to perform
>> calculations in more common way?
>> Thank you!
>> R-help at r-project.org mailing list
>> PLEASE do read the posting guide http://www.R-project.org/posting-
>> and provide commented, minimal, self-contained, reproducible code.
> R-help at r-project.org mailing list
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
More information about the R-help