statistical model selection (and Rosa D)

Lunnan & Hjort brit.lunnan at chello.no
Mon Mar 22 03:59:40 CET 2004


Hans Kiesl (from the University of Bamberg) asks me: 

<< Leafing through your papers in the December 2003 issue 
of the Journal of the American Statistical Association, I noticed 
the following reference:

Hjort, N.L., and Rosa, D. (1999), "Who won?", 
Speedskating World 4, 15-18.

I guess the article is about the influence of a skater's last track 
(inner vs. outer track) on the outcome of speedskating competitions; 
this is of course completely off topic here. What I'd like to know, 
though, is Mr. "Rosa, D."'s contribution to the paper (and whether 
it makes this posting on topic). >>

Thanks for noticing, and for looking at these articles in the JASA.
I'm very proud of them, and of the many discussion contributions
that followed. I was also happy to be able to bring both 
speedskating and a connection to Mr Rosa D onto the gravely 
serious pages of JASA, devoted here to mathematical discussions 
of statistical-probabilitistical model selection and model averaging 
methodology. It was to clarify the sometimes conflicting roles
and uses of statistical regression models in applied scientific practice
that I pointed to my speedskating work (causing the Olympic
rules to be changed, from Nagano 1998 onwards), and where 
one of the related points is being nicely illustrated in the mentioned
Hjort & Rosa publication. It turned out that Mr Rosa D has a good 
grasp of certain engineering aspects of the problems that sometimes
conspire to confront speedskaters.

Yes, this should qualify as on-topic, I believe, and the serious
Rosa D reader should not be without a copy of the above publication.
I think *some* copies might still be had from the Dutch editor of 
SpeedSkating World, Irene Postma, also an occasional reader of 
the DCML pages. She might be contacted at "irenep at bos.nl" . 

(And no, the direct sample average is not always the most efficient 
estimator of the mean of an unknown distribution. For the double
exponential family, for example, the standard deviation of the 
sample mean is sqrt(2) bigger than that of the median, for large n.)

Nils Lid Hjort 





More information about the DCML mailing list