[Corpora-List] Summary (Fisher Exact tests)

[Stefan Th. Gries]

Hi everybody

Nearly three weeks ago I posted the following query to this list:

-----------------------------
For a current project involving collocations, a colleague and I need to
compute several hundreds of F-E tests, some of which involve very high
marginal totals even though the cooccurrence frequencies are very small. The
question now is, do you happen to know of any
(Windows/Macintosh/Linux-based)software that can cope with the enormous
sizes of the resulting figures? An example:
2    3    5
1,584    10,204,711    10,206,295
1,586    10,204,714    10,206,300

Some scripts on the web seemed to do the job, but on second looks, the
results seemd wrong because transforming tables along the main diagonal)
with such high figures lead to different results, inviting the inference
that the scripts only computed approximations ... Any idea(s)?
-----------------------------

I thank the following people for their advice:
Christer Johansson, who suggested programming the test in MATLAB or MAPLE;
Alexander S. Yeh, who directed me to versions of Lisp (like Allegro common
lisp, I believe, but NOT the lisp used to implement the EMACS editor) which
can handle integers with large (infinite?) numbers of digits;
Marc Feeley, who proposed the Scheme programming language (see
http://www.schemers.org/Documents/FAQ/#implementations);
Richard <z.xiao at lancaster.ac.uk>, who suggested to use SPSS.

In the meantime, I have also myself also come across R (see
http://www.stat.ufl.edu/system/man/R/doc/html/), which is based on Scheme
and also seems capable of handling the extremely large numbers mentioned
above.

My thanks to all who contributed and best wishes for a happy new year

Stefan

Stefan Th. Gries
-----------------------------------------------------------
IFKI, Southern Denmark University
http://people.freenet.de/Stefan_Th_Gries
-----------------------------------------------------------