[Corpora-List] Which Statistical Test is Suitable

[chris brew]

>
>
> CB
>
>  But, if you do manage to set up sufficiently precise hypotheses,
>> and associate numbers with the hypotheses, statistical reasoning
>> definitely can help.
>>
>
> I agree that statistics can help.  But there are many models for
> generating statistics.  Should you give higher weights to typing
> mistakes, dictionaries, legislation, or common usage?
>
> John


Yes indeed. These are the questions. I have no idea how to decide. The point
I was trying to make, which I believe most of us agree on, is that
statistical methods are just a tool for reasoning about uncertainty, so the
burden falls back on the researcher to make sensible choices about how to
model the situations that are of interest. This fundamental fact about what
statistics is can often be obscured by practice and pedagogy in psychology
(and probably also applied linguistics), because statistical methods are
used mainly in stereotyped experimental designs for which well-researched
statistical tests are firmly established, with most of the conceptual kinks
already ironed out by higher authority. In that light it can seem that the
task reduces to finding the "right" statistical test. I think that this
attitude is dangerous and misguided in any field, and that you need a deeper
understanding of what you are doing in order to draw safe conclusions. For
problems as conceptually tricky as the original poster's, there is no
immediate reason to think that any standard statistical test will be even
close to right.


-- 
Chris Brew, Educational Testing Service
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20110713/ec50e4b9/attachment.htm>
-------------- next part --------------
_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora