A few years ago, on this blog and at Climate Audit, there was a healthy discussion of a paper by Greg Holland and Peter Webster that claimed definitive attribution of hurricane activity to greenhouse gas emissions (PDF). Now a paper by Sim Aberson is out in the current issue of BAMS (PDF) which uses the Holland/Webster paper as a good example of how not to do statistics.
The Aberson paper is summarized as:
A cautionary tale in which previously published results are shown to be invalid due to the lack of statistical analyses in the original work.
Eric Berger, of the Houston Chronicle, covers the paper on his blog, and here is an excerpt:
Aberson is basically saying that the statistics underlying Holland's arguments in the 2007 paper are off. Way off.
So bad, in fact, that the way in which Holland draws his conclusions about the relationship between Atlantic sea surface temperatures and hurricane activity could also be used to conclude there's a meaningful relationship between Atlantic sea surface temperatures and five-year running means of:
- The number of Republican Party members in the U.S. House
- The number of years since the crowning of a new pope
- The number of games the New York Yankees won in the current season
- A random number between zero and one
If that weren't bad enough, Aberson concludes his paper with the following lecture:
The clear need for timely scientific results should not be a reason for shortcuts in the scientific process; correct statistical analyses must be performed to determined the likelihood that the hypothesis tested is valid.
Berger says that he doesn't know who is right in the debate. I agree with Aberson, Holland/Webster is a good example of how not to do statistics.
