The following article is from Dr. John Gardenier, concerning the ethical use of statistics. You are invited to read it and join the discussion on it.
 

Status of Statistical Ethics, 1996

Paper presented at the Joint Statistical Meetings in Chicago, IL,
August, 1996
Author's e-mail address: jsg6@cdc.gov.
        

                WHAT AND WHERE ARE STATISTICAL ETHICS?
        
John S. Gardenier, Centers for Disease Control and Prevention, National
Center for Health Statistics, 6525 Belcrest Road, Room 1120, Hyattsville,
MD 20782

Keywords: Ethics, Statistics, Professionalism

Abstract

There exist several "codes" or "guidelines" which purport to
guide ethical statistical practices.  All have varying degrees of scope,
specificity, and topical content.  There are ethical issues of
statistical interest which are not specifically addressed in any of
them.  There are issues of common ethics, of scientific ethics in
general, and of professional ethics, such as university professors'
ethics, which may have direct relevance to some statisticians, but are
tangential to others.

Within a single society, such as the American Statistical
Association (ASA), there are multiple loci of "official" ethical
pronouncements, which are nowhere collated into any single document.
Furthermore, the one document which specifically purports to state the
society's position was defined several years ago as a work in progress.
It has not progressed. If an ASA member is asked to define what
constitutes statistical ethics, the answer might reference ASA's
"Ethical Guidelines for Statistical Practice."  Yet, that document does
not incorporate some of the ethical principles formally adopted by the
Society and reflected in various official publications.

This state of affairs leads the author, currently Chair of the
ASA Committee on Professional Ethics, to several questions.  What are
statistical ethics?  What should be taught as statistical ethics?  To
whom should statistical ethics apply?  Above all, what if any action is
required?


Background: Origins of "Ethical Guidelines for Statistical Practice"

The existing guidelines were formulated in 1983 after much
discussion among a drafting committee, the Board of Directors, and the
membership in general.  A good overview of the thinking at that time can
be found in The American Statistician, February 1983.  In 1989, some
administrative verbiage was removed, but the Guidelines did not change
in any material way.  They remain unchanged today.

In 1983, fourteen reviewers offered varying degrees of praise
and criticism.  Positive reviewers said it was an excellent start and
would benefit from further development.  Less kind commenters used words
like "platitudinous," "bland," and "deficient."  The overall thrust of
the reviews was that these general guidelines, perhaps with or without
changes, should be backed up by more specific rules, case studies, or
guides to action.  It was not clear whether the Guidelines were
addressed to ASA members only, to all professional statisticians, or to
all practitioners of statistics, regardless of their professional title.

In terms of the categories of such documents: aspirational,
regulatory, or educational; there were firm objections to regulatory
rules.  These guidelines were seen as largely aspirational.  The
drafters' reply to the reviewers agreed.  They stated that the goal, not
originally explicit, was an educational code, adding: "We hope that the
Ethical Guidelines will evolve in this direction."

The ASA Committee on Professional Ethics, established after a
three-year trial, was to establish liaison with other groups, to promote
discussion of ethical issues, and to recommend actions to the Board of
Directors to modify or augment the Guidelines.  Those who are satisfied
with the Guidelines and those who strenuously object to them both use
the same argument.  The current guidelines provide little or no
practical basis for determining whether specific actions of practicing
statisticians are ethical or unethical.  Those in favor feel the working
environments and application areas of statisticians are too varied to
conform to common specific guidelines; only general ideals are broadly
applicable.  Those who object feel the Guidelines are somewhat
hypocritical, providing only general "lip service" to the concept of
ethics, but avoiding many of the real life problems faced by
statisticians.

The world of professional ethics is moving on. The American
Association for the Advancement of Science has an active Professional
Society Ethics Group and a newsletter.  The DHHS Commission on Research
Integrity has suggested a revised definition of and approach to research
misconduct.  There are frequent articles on research ethics in Science
and Nature as well as in the general press.  A new journal premiered in
1995, Science and Engineering Ethics.


Whose Statistical Ethics?

Certainly, ASA is not the only source of statistical ethics.
Here are a few of the other professional societies involved with
statistics which have ethics documents:

International Statistical Institute
Council of American Survey Research Organizations
American Federation for Clinical Research
Educational Testing Service
International Epidemiological Association
Pharmacoepidemiologists
American Sociological Association
American Psychological Association


Dispersed Ethics Functions in ASA

Completely apart from either the "Ethical Guidelines for
Statistical Practice" and the Committee on Professional Ethics, there
are numerous sources of official ASA ethical responsibilities and
pronouncements, each known only to a minority of the ASA membership.  A
review of the charters of sections and committees reveals, inter alia:


The Biopharmaceutical Section addresses:

Evaluation of drug and device safety and efficacy, Health promotion and
disease prevention and intervention activities, Impact of federal policy
on human and animal research, Standards of design, evaluation, and
reporting of biochemical, biological, human, and animal health
experimentation, Ethics related to human and animal experimentation, and
Continuing education programs, including but not limited to statistics for
statisticians and other researchers working in these areas.

The Section on Statistical Consulting includes among its objectives:

   Promote the development and use of ethical standards in consulting, and
   Promote the professional stature of consultants and statistical
   scientists in general.

The Section on Statistics in Marketing includes the objective:

   "The Section will cooperate with other professional organizations to
   advance statistical science in marketing, stimulate research, and
   promote high professional and ethical standards."

The Section on Survey Research Methods (SRM) includes among its interests:

   Ethics relating to the conduct of survey research and to respondents
   and potential respondents.

One of the SRM functions is:

   To develop and publicize standards for the design, conduct, and
   reporting of surveys, and for the assessment and evaluation of survey
   practices. 


Formal Resolutions of the ASA Board

All official ASA positions must be decisions of the Board of Directors.
The Board adopts positions in various ways: by approving publications
prepared by component Sections or Committees or by staff, by recording
minutes of Board meetings at which decisions are made, and by adopting
formal policy statements.  Among the latter statements are the following
ones which pertain to ethics:

Guidelines Against Discrimination Because of Sex, Guidelines Supporting
Minorities in Statistics, Endorsement of Resolution of the American
Association for the Advancement of Science (AAAS) on Discrimination
Against Sexual Minorities (recognizes their contributions and rejects
biases against them which are unrelated to their professional
qualifications), Guidelines Against Sexual Harassment Resolution on
Adherence to Traditional Academic Standards in Granting of Academic
Degrees (revocation only "upon discovery and verification of fraud or
misrepresentation in the work offered for the degree.") Adoption of
Research Industry Coalition Policy on 900 Numbers (calls for disclaimer
statements in reporting the results of polls which are not statistically
representative of any defined population.)



W. Edwards Deming

Statisticians proudly claim "Ed" Deming as one of our heroes.  He helped
revolutionize industrial quality and productivity.  He was also a stickler
for the ethical practice of statistics.  Among items featured in his
personal code of conduct, which are not found in other statistical ethics
sources: 


Devote oneself to the statistics profession alone.  

Set ethical rules with the client up front.  

Limit own role as statistician only to those functions supported by
statistical theory. 

Draw conclusions only about the frame, not the population. 

Retain exclusive publication control over statistical matter one reports. 

Assure reports clearly state who did what in conducting the study and
document all meaningful possible sources of error. 

Prepare own expert testimony.  (Lawyer can help clarify, but not
originate, positions.)

Reserve right to break off an engagement without explanation.  Do so if a
client or colleague (on a joint study) does unacceptable work. 

Retain all rights to statistical theory developed on an assignment to
share them with the profession. 



Missing Pieces

With many diverse sources helping to define statistical ethics, one would
think that all the issues were covered among them.  Not so. Other issues
which arise in statistics as well as in other areas of research and
application include: 

Authorship standards, Standards of promotion and tenure, Whistle blower
protection, Criminality or moral turpitude, Conflicts of interest,
Deception for research purposes, Considerations in special populations,
such as children or the retarded, and Relationships with colleagues in
joint studies. 


What Are Statistical Ethics?

Given all of the above, how are practitioners or students to
define statistical ethics?  Some have advocated limiting formal
statements of statistical ethics only to issues which are peculiar to
the statistics profession.  That is hard to define; is the statistics
profession hurt, for example, by discrimination against, or harassment
of, statisticians due to sex, race, or sexual orientation?  Clearly, ASA
Boards have felt, at the least, individual statisticians are hurt that
way.

The argument has been raised that issues covered either by law
or by institutional regulations should be declared "out of scope" for
statistical ethics, which would deal with remaining issues.  At least
one statistician has commented that there ARE no remaining issues at his
institution.  It has rules on everything.

Are there standards for the content of ethical codes which could
be referenced in defining statistical ethics?  No.  That is a further
complication.  Statisticians often must work on interdisciplinary teams.
They face problems when all colleagues are subject to ethical codes or
guidelines, but these may be somewhat different for each participant.

Ethical issues are frequently intertwined with issues of law and
politics.  Who has the authority to set the local ground rules for how a
specific study will be conducted?  Can and will higher authority
intervene or override decisions?  Are tricky legal issues involved?
Under what circumstances should ethical statisticians throw up their
hands and leave decisions to the lawyers?  Whose lawyers?

Jaffe and Spirer have commented, "Have we ourselves misused statistics? 
We hope not, but as you will see, it is hard to avoid an inadvertent
misuse and we are as fallible as any other statistician."  Some
statisticians argue for a separation of ethics and competence.  Well,
honest disagreement about methods is ethical, of course, but we sometimes
see abuses which exceed reasonable tolerance. 


What Should We Teach Students?

Because of the full complexity of ethical issues in a large,
diverse universe of contrasting environments, applications, and
collaborative structures, some feel that the only tractable problem is
deciding how to teach statistical ethics to students.  After all, this
cannot be expected to deal with all possible issues.  It must be
simpler, right?

Yes and no.  Some new issues arise in teaching, as opposed to
defining, statistical ethics.  First, some insist that no student can
truly understand ethics, as opposed to memorizing enough words to pass a
test, without a grounding in some philosophical theory of ethics.  This
leads to discussions of whether some philosophy course should be a
prerequisite to an ethics course or whether a philosophy professor might
be a guest lecturer in a statistical ethics course.

There are others who feel a separate course in ethics is futile.
Worse yet, such courses can leave students with the impression that
ethics is a separate "thing" from "doing statistics," not part of the
actual work.  Others claim success with separate ethics courses.

There are laboratory courses in science which are graded on: the
results, the neatness of the notebook, and the clarity and accuracy of
the report.  Makes sense?  The problem is that experience has shown that
students get the expected results and look good even when essential
laboratory equipment, unbeknownst to them, has become miscalibrated or
reagents are defective.  In short, the ethical principle the students
are really learning is to provide the desired results regardless of what
happens in the actual experiment or data gathering.

To counter this, some authors advocate embedding ethical issues
into every aspect of the teaching of technical material.  They do not
even support reserving one or a few lectures on ethics within a
technical course.  There is some experience of the main professor
failing to show up for guest lectures on ethics.  Those professors
apparently fail to understand that such behavior in itself is a more
powerful, and adverse, lesson to the students than anything the guest
lecturer could say.

Teaching statistical ethics is not necessarily simpler than
defining the term for the full range of practitioners.  Good solutions
depend on the wisdom, dedication, and enthusiasm of individual teachers.
No one approach will fit all teachers or classes.

Statistical Ethics for Whom?

It seems reasonable that ASA can only set ethical standards for
its own discipline.  We have little to say about how sociologists,
econometricians, engineers, demographers, astrophysicists,
psychometricians, medical doctors, or epidemiologists (to name just a
few) apply statistics themselves.

Actually, ASA cannot even "set" ethical guidelines for its own
members.  It cannot enforce them; it can only reach for consensus.
Furthermore, not all professional statisticians are members of ASA or of
the affiliated societies represented at the Joint Statistical Meetings.

Arguably, much more statistical research is done without the
participation of professional statisticians than with them.  To limit
the scope of statistical ethics to "statisticians" is to abandon most of
what is done in the name of the discipline.


Required Action

The "Ethical Guidelines for Statistical Practice" must be
updated.  The current document does not represent statistical ethics in
general or even within ASA.

The ASA Committee on Professional Ethics can accept
responsibility for coordinating the effort.  Many other ASA officers,
representatives of affiliated societies, and interested parties from the
general ASA and affiliates memberships, need to participate if the
result is to be valid, general, and credible to a sizable portion of the
community of statistical practitioners.

Once updated, the ethical guidelines should be available as
widely as possible so as to encourage sound and ethical statistical
practice generally, not only by statisticians.


References

1.  An article and several comments by various authors appeared in The
American Statistician, Vol. 37, No. 1, February, 1983.  Specifically:

Ellenberg, Jonas H., "Ethical Guidelines for Statistical Practice: A
Historical Perspective," pp.1-4.

Martin, Margaret E. Comment, pp. 7-8.

Gehan, Edmund A. Comment, pp. 8-9.

Rice, Dorothy P. Comment, p. 9.

Mosteller, Frederick, Comment, pp. 10-11.

Boardman, Thomas J., Comment, pp. 11-12.

Bross, Irwin D. J., Comment, pp.12-13.

Kempthorne, Oscar, Comment, p. 14.

Remington, Richard D., Comment, pp.14-15.
Solomon, Herbert, Comment, p. 15.

Greenhouse, Samuel W., Comment, pp. 15-16.
Kish, Leslie, Comment, p. 17.

Free, Spencer M., Comment, p.17.

Roberts, Harry, V., Comment, p. 18.

Barabba, Vincent P., Comment, pp. 18-19.

2.  Professional Ethics Report, a newsletter of the American Association
for the Advancement of Science, Scientific Freedom, Responsibility, and
Law Program.  (Washington, DC: AAAS, quarterly.)

3. DHHS Commission on Research Integrity, Integrity and Misconduct in
Research.  (Washington, DC: Department of Health and Human Services,
Office of Research Integrity, 1995.)

4. Stephanie J. Bird and Raymond Spier (eds.), Science and Engineering
Ethics (Guilford, Surrey, UK: Oragen Publications, quarterly.)

5. W. Edwards Deming, "Code of Professional Conduct," International
Statistical Review, Vol. 40, No. 2, 1972, pp. 215-219.  Also reprinted
as a brochure for his consulting practice.

6. Whicker, Marcia Lynn and Kronefeld, Jennie Jacobs, Dealing with
Ethical Dilemmas on Campus.  (Thousand Oaks, CA: SAGE Publications,
Inc., 1994)

7. Wang, Chamont, Sense and Nonsense of Statistical Inference:
Controversy, Misuse, and Subtlety.  (New York: Marcel Dekker, Inc.,
1993, p. v.)

8. Jaffe, A. J. and Herbert F. Spirer, Misused Statistics; Straight Talk
for Twisted Numbers.  (New York: Marcel Dekker, Inc., 1987, p. vi.)