Re: cramer-von-mises

jeremie <njphoto21@xxxxxxxx> wrote in
news:45d486a3$0$19051$426a74cc@xxxxxxxxxxxx:


David Winsemius a écrit :
jeremie <njphoto21@xxxxxxxx> wrote in
news:45d377f1$0$377$426a74cc@xxxxxxxxxxxx:

Hi folks,

I have found contradictory information on the Cramer Von Mises
statistical test regarding the critical values to use:
1.
http://www.weibull.com/RelGrowthWeb/Critical_Values_for_Cramer_von_Mi
se s_Test.htm 2. http://rkb.home.cern.ch/rkb/AN16pp/node45.html

In 1, for a sample of size 30 the critical value is 0.33
In 2, for a sample of the same size, the critical value for W^2 is
0.743/30=0.0247667.

Do you have any hints on the one I should beleive?

Could it be that you are comparing (1) which is a specific
application of the CVM to the case of one-sample data versus an
assumed Poisson distribution to (2) which is a general result for any
distribution? You should know the assumptions underlying the actual
test statistic and assumed distribution that you are using. There is
not just one CVM test, but many versions, just as there is not just
one "chi-square test".

Here are some examples you may find entertaining:

http://stinet.dtic.mil/cgi-bin/GetTRDoc?AD=ADA262554&Location=U2&doc=G
etTRDoc.pdf

http://www.hindawi.com/GetPDF.aspx?doi=10.1155/BSB/2006/85769

Thank you david,
I am trying to test, kind of automatically, if sets of data I have can
be fitted with some known laws such as Pareto and exponential.
For the moment, I am using (2) but it seems that the critical values
are too low which makes the test be too strict (almost all my tests
fail but I can visually see that my curves are not that far from an
exponential), I would need something more "flexible". That is why I
was wondering if the critical values I am using are the good ones.
Do you have advices for me?

You will need to hope that a real probabilist chimes in. I doubt that I
will have the answers for you. In your effort to generate more interest,
it might help if you described your methods, data situation (types of
data, numbers of data points, etc) and goals in more detail. I for one,
am not even sure I know what you mean when you say your "tests fail."
Fail to give the the expected answer, or fail to reject the hypothesis of
similarity? If the critical values are low, then it would seem that you
would be getting a lot of rejections, which many people would not call
"failing" but rather "working".

--
David Winsemius

.

Relevant Pages

  • Re: Question for Alan Shank...or others
    ... How far off from expectation do I have to be to fail the test? ... We perform some statistical test, and it fails at the 99.9% ... He happened to record every outcome from the past 2000 outcomes ... But the simple fact that we are looking backwards ...
    (rec.gambling.craps)
  • Re: Splitting samples to minimize false positives?
    ... The sample was taken from the expected distribution ... that sample will fail also. ... Say that the new db doesn't have a data entry issue. ... a proportion test will fail yield a 5% false positive ...
    (sci.stat.math)
  • =?UTF-8?Q?Re:_Monte_Carlo_NEWS_=E2=80=93_1971._JASA?=
    ... David Winsemius, you know very well and teach fully Jack Tomsky that is elementary the following: ... We accept (fail to reject) H0 BUT WE NEVER KNOW IF THE HYPOTHESES IS TRUE. ... I am sure that this problem or equivalent ones, are used in your Country. ...
    (sci.stat.math)
  • Re: Question for Ray Martinez
    ... referred to "centers of creation" when describing the distribution ... of species. ... fail or be modified. ...
    (talk.origins)
  • finite divisibility in probability
    ... A probability distribution F on the real line is infinitely ... numbered cumulant is negative, infinite divisibility will fail). ...
    (sci.math.research)