Date: Fri, 10 Nov 2000 20:09:01 +0000
Reply-To: David Kreil <kreil@EBI.AC.UK>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: David Kreil <kreil@EBI.AC.UK>
Subject: Corrections for tests of Poission distribution with unknown
parameter
Content-Type: text/plain; charset=us-ascii
Dear Colleagues,
I would be most grateful for any help with the following question (and
I would greatly appreciate for a cc of your reply to "kreil" at
"ebi.ac.uk")
With many thanks for your support,
David.
-------
I tried to assess whether several datasets are distributed normally
(each), which some of them (with a large mean) seem to be. The others
rather appear to be Poisson distributed (with a mean between 0 and 2).
Obviously, for very large means, the two distributions would be
similar.
I have been using the SPSS Kolmogorov-Smirnov one-sample test for
comparing the empirical distribution of each data set to the Poisson
distribution. Yet, as I understand it, the statistics change if the
parameters of the Poisson distribution are derived from the data set
to be tested.
Indeed, comparing the SPSS Kolmogorov-Smirnov one-sample test for
Normal distribution with Normal distribution parameters derived from
the data, I find results different to the tests for Normality used in
the SPSS Explore data analysis (Kolmogorov-Smirnov with Lilliefors
corrections and Shapiro-Wilk).
Now, are there any corrections to the K-S 1-sample test for Poisson
distribution that should be applied when the Poisson parameter is
derived from the test data?
Are there any other tests I could use that do not have this problem?
-------
---------------------------------------------------------------------------
David Philip Kreil at the EMBL ("`-''-/").___..--''"`-._
European Bioinformatics Institute `6_ 6 ) `-. ( ).`-.__.`)
Wellcome Trust Genome Campus (_Y_.)' ._ ) `._ `. ``-..-'
Cambridge. Kreil@EBI.ac.UK _..`--'_..-_/ /--'_.' ,'
++44 1223 49-4663, fax -4468 (il),-'' (li),' ((!.-'
