[Corpora-List] collocations and exact hypothesis tests

[Daniel Wiechmann]

Dear all,

I have a tiny question concerning collocates and their statistical 
associations. I believe that exact hypothesis tests, like Poison or Fisher, 
are among the most reliable tests around to express degrees of association, 
so I have been using Fisher's exact test quite a lot. I trust that it is one 
of their merits to deliver reliable results regardless of the sample size. 
However, association scores derived from different samples may not be 
comparable due to that test's sensitivity to different sample sizes. In 
order to allow sensible comparisons of association scores derived from 
different samples, I have now turned to (a discounted version of) the log 
odds ratio to express the degrees of association.

But maybe this isn't really necessary...can anybody help me out and comment 
on the Fisher's exact tests sensitivity to sample sizes?

Any help would be greatly appreciated.

Best,
--Daniel

--------------------------------------------------------------------

daniel wiechmann
department of british and american studies
linguistics: language and cognition
friedrich schiller university, jena

www.daniel-wiechmann.eu

--------------------------------------------------------------------