The standards of medical statistics

Athel athel at IR2CBM.CNRS-MRS.FR
Mon Feb 17 05:21:10 EST 1997


Vassili P. Leonov wrote:

"I have collected large collection of examples of misuse of statistics
in the dissertations and articles. Their analysis shows, that most of
all errors in use of statistics is observed in medicine and biology.
Especially it is bad when it is observed in medical work. Errors in use
of statistics in medicine result in errors in methods treatments of the
ill people.

"I shall be grateful for any information how is solved this problem in
USA, UK and as a whole in WHA."

There is no doubt that this is a very serious problem, certainly in
biology (I can't speak for medicine), and it is not confined to Russia.
So I would say that it is *not* solved in USA, UK ...

You can find examples of misuse and misunderstanding of statistics at
all levels, in journals from the most famous to the most obscure. Part
of the problem is that many journals require (or authors think they
require, which often comes to the same thing) conclusions to be
supported by statistical tests, but they don't require any evidence that
authors understand the tests they are using. So what happens? people use
cookery-book statistical tests in a mechanical fashion, without checking
whether the assumptions implicit in the tests are applicable to their
data, or even in many cases realizing that there are any assumptions
implicit. For referees to check carefully whether tests are appropriate
and have been calculated correctly is a lot of work, and often this is
not done.

I am pessimistic enough to doubt whether any solution is in sight, but I
would suggest to Dr Leonov that he draw attention to the examples he is
finding -- either here on the net, or in articles submitted for
publication, in the hope that he may influence a few readers to take a
better direction. For myself, I have tried in my book "Analysis of
Enzyme Kinetic Data" (http://193.50.234.64/lcbpage/athel/leonora.html)
to preach the message that statistical methods that the user understands
(at the simplest level, graphs; at a slightly more sophisticated level,
distribution-free statistics) are nearly always preferable to
potentially more powerful techniques that are either not understood or
imply assumptions that do not apply to the problem at hand (or, often,
both).

Athel

Home page: http://ir2lcb.cnrs-mrs.fr/lcbpage/athel/home_page.html





More information about the Bio-soft mailing list