James Annan wrote:
> Raymond Arritt wrote:
>
>> More broadly, please see "The Insignificance of Significance Testing"
>> by Neville Nicholls, Bulletin of the American Meteorological Society,
>> Vol. 82, No. 5, pp. 981–986 (2001).
>>
>> Public access to BAMS is free. Go to:
>>
>> http://ams.allenpress.com/amsonline/?request=get-toc&issn=1520-0477&volume=82&issue=5
>>
>>
>>
>> and scroll down (under "commentary and analysis").
>>
>
> This paper repeats much of the well-known (but frequently ignored)
> problems with statistical hypothesis testing. It should be required
> reading for all climate scientists.
>
> However, it unfortunately drops a clanger of its own.
>
> It recommends generating confidence intervals (rather than just a
> p-value for the null hypothesis):
Sigh. Dog ate my homework. Try again.
"The reporting of confidence intervals would allow readers to address
the question 'Given these data and the correlation calculted with them,
what is the probability that H0 is true?'".
However, this is false, and this error (equating a frequentist
confidence interval with a bayesian credible interval) is equally as
pernicious as the simple prosecutor's fallacy that Nicholls rails
against. Indeed, it is essentially the same error, of interpreting
P(D|H) as P(H|D)! This mistake seems well-nigh ubiquitous in climate
science, eg the whole of the detection and attribution field (chapter 12
in the IPCC TAR) falls into this trap, and MBH's "likely the warmest"
does the same. I noted the same thing in a recent D&A paper (which is
co-authored by some of the most prominent names in the field) a few days
ago and am awaiting their explanation/excuses...
Nicholls cites Wilks (Statistical methods in the atmospheric sciences)
for confidence interval calculation, but Wilks makes the same error.
Unfortunately, I think my chances of persuading the climate science
community of the error is rather slim given such authority lined up in
opposition. Nevertheless, it is trivial to generate confidence intervals
that are entirely valid and correct in the frequentist sense, but which
cannot possibly contain an unknown parameter with the specified
probability.
Of course, I'm happy to acknowledge my own fallibility and would be
interested to hear of any arguments to the contrary...
James
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
Global Change ("globalchange") newsgroup. Global Change is a public, moderated
venue for discussion of science, technology, economics and policy dimensions of
global environmental change.
Posts will be admitted to the list if and only if any moderator finds the
submission to be constructive and/or interesting, on topic, and not
gratuitously rude.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at
http://groups.google.com/group/globalchange
-~----------~----~----~----~------~----~------~--~---
