>
> Post by Steve Hilliard
> >>First a "thank you" to all of you who set me straight regarding the
> intimidating K-S test. Several mentioned that B-D software provides
> this test--do any other packages offer it? I'll have to check it
> out,<<
>
> Hi Steve
> The K-S test is also available in StatGraphics (for Windows) - probably
> also in other Stat packages. However, I got to thinking about it after
> trying to explain it to a colleague - works on differences in cumulative
> frequencies - and I realised that I would like to know the nuts and bolts
> of the method instead of just accepting the package version (I used to
> write my own stat software in the old days - from textbook formulae, etc).
> I trolled through all the statistics books I could find in a nearby
> reputable student bookstore - only one mentioned it - and only as something
> the reader might wish to go on to. I did find a table of probability
> distributions for it. I'll probably have to pay a visit to the university
> stats department for info.
> Does anybody out there have a detailed description - or reference (text or
> paper) to the Kolmogoroff-Smirnov test - solutions and probabilities.
> Please e-mail me if you do - my curiosity is irked.
> Robin Barclay
> SNBTS, Edinburgh
>
Hi Robin, hi Steve (and to all of you interested in "sadistics")
First of all: Yes, Robin is right, the K-S-test is included in some
other statistics-packages. We here use the StatSoft-Statistica
package (although we have also written some software on our own).
Anyhow, I had a look at the handbook, but all I could find out about
the K-S-test was that it is used here to check whether a respective
distribution is normal or not.
Anyhow, we (the stats-experts of our group and me) agree with
Robin: there's always the problem if you relay on stats-packages,
esp. if the answer of these is only "there is a significant
difference" or "there is no difference".
This answer may be nice and easy, but there's a lot of possible
errors in it, esp. if you don't know what's behind these test and
how they obtain these results.
All in all, we think that the first question ("a test for
statistical significance between overlapping populations") isn't
easy to accomplish and the K-S-test is only one possible way.
Another way is to check the significant difference between these
populations.
I know that this answer will not be satisfying for Steve, but from
our point of view there isn't an "easy" to use statistic.
Hope this contribution did help anyway
Best regards
w.w.
Mag. Wolfgang Wlach Institut für Pharmakognosie
------------------- der Universität Wien
wolfgang.wlach@univie.ac.at Pharmaziezentrum
Tel. +43/1/313 36/8270 Althanstr. 14
Fax. +43/1/313 36/772 1090 Wien
This archive was generated by hypermail 2b29 : Wed Apr 03 2002 - 11:50:21 EST