Misconduct, Especially Distortion of Uncertainty

Tore Schweder wrote:

Dear Dr Gardenier,

I just received the May issue of Amstat News, where Draft for General
Comment (Ethical Guidelines for Statistical Practice) appears. I have
two main points to make.

1. A reference to

National Academy of Sciences. 1992. Responsible Science; Ensuring the
Integrity of the Research Process, volume 1-2. National Academy Press,
Washington , D.C.

would seem in place. The two volumes are most readable, and they
provide sound definitions and guidelines. Your proposed guidelines
would be stronger if they were build upon the more general guidelines in
the Academy report. You have a point II.G. on "Responsibilities
regarding allegations of misconduct". The concept of misconduct in
science is, however, in the Draft only implicitly defined by the (in my
view too vague) specification of proper conduct. The Academy report is
much more explicit and up-front. They list three basic types of
misconduct: fabrication, falsification and plagiarism. To
statisticians, in science and outside, fabrication and falsification of
data are of course the worst of sins. However, such sins can be
conducted in subtle ways, even unintentionally, and I fear that they are
committed more often than we wish to think. Your points I.C.2. and
II.A.1. touch upon this, but in my view all too softly. Types of
falsification and fabrication needs to be spelled out in the Guidelines
or in a supporting documents. Case studies would be most welcome
reading. Case studies are consistent with your point II.G.1. It is
painful to flush out misconduct, but it is necessary. What we see these
days of what has been going on inside big tobaco research is telling.
Do ASA have members in big tobaco. Have they conducted ethically?

2. In addition to fabrication and falsification, I suggest that there is
a third category of misconduct that deserves attention, particularly
among statisticians. That is "distortion of uncertainty". Concepts for
understanding uncertainty and methods for handling it is the core of our
trade. We have therefore particular responsibility for unbiased and
honest assessment and reporting of uncertainty. In the big tobacco
case, I fear that scientists (statisticians) withheld (were forced to)
data and reports in the interest of not reducing uncertainty to
appropriate levels.
Uncertainty can be distorted both ways. Expanding uncertainty with the
intent to decieve might be done for one reason (is it going on with
respect to global warming?), while improperly diminishing uncertainty
can be done for opposing reasons. Both activities are bad, and
statisticians need to be alerted to such pitfalls and their potential
consequences.

As you can guess from the above, I have an interest in these matters. I
have studied science in the management of whales, and have found that
gross misconduct was carried out - from both sides, of the category
uncertainty distortion (expansion). My paper on this was, of course,
not well received from those who currently carries out this activity,
and whom I name. And they have been forceful enough to prevent the
publishing my paper. They actually threatened with libel suit the
publisher of a journal that had accepted my paper for publication,
making the management of the journal override the editor's decision and
preventing publication.

You might post a summary of my comments on your web site.

Tore Schweder
Professor of Statistics
Department of Economics
University of Oslo, Norway

(Member of ASA, IMS, ISI, Norwegian Academy of Science)

Response from Dr. Gardenier:

Thank you, Tore. I am inclined to include the National Academy of
Sciences reference. However, the definition of misconduct as (limited
to) fabrication, falsification, and plagiarism (FF&P) is wholly
inadequate. By omission it implicitly condones a much larger range of
serious misconduct, including the deliberate misrepresentation of
uncertainty and the suppression of legitimate scientific debate. No one
definition can apply to all statistical practitioners in all
environments and applications. The term "misconduct" must be left vague
in the Guidelines and be subject to relevant interpretation by different
readers. Otherwise, to the extent your comments would toughen the
guidelines instead of limiting them, our Committee will consider them
carefully.