-
This article was downloaded by: [TIB &
Universitaetsbibliothek]On: 03 January 2015, At: 08:49Publisher:
Taylor & FrancisInforma Ltd Registered in England and Wales
Registered Number: 1072954 Registered office: Mortimer House,37-41
Mortimer Street, London W1T 3JH, UK
The American StatisticianPublication details, including
instructions for authors and subscription
information:http://www.tandfonline.com/loi/utas20
Statistics and EthicsStephen B Vardemana & Max D Morrisaa
Stephen B. Vardeman is Professor, and Max D. Morris is Professor,
Department ofStatistics and Department of Industrial and
Manufacturing Systems Engineering, IowaState University Ames, IA
50011-1210 . The authors gratefully acknowledge the generousinput
of a number of colleagues. Karen Kafadar, Bob Stephenson, Bill
Meeker, and DeanIsaacson provided detailed comments on a first
draft of this article. And the input of KenRyan, Tammy Brown, Mike
Moon, Bill Notz, Tom Dubinin, Frank Peters, David Moore,
BillDuckworth, Bruce Held, Dennis Gilliland, Bobby Mee, Doug
Bonett, Dan Nettleton, and HalStern is also gratefully
acknowledged.Published online: 01 Jan 2012.
To cite this article: Stephen B Vardeman & Max D Morris
(2003) Statistics and Ethics, The American Statistician, 57:1,
21-26,DOI: 10.1198/0003130031072
To link to this article:
http://dx.doi.org/10.1198/0003130031072
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy
of all the information (the Content) containedin the publications
on our platform. However, Taylor & Francis, our agents, and our
licensors make norepresentations or warranties whatsoever as to the
accuracy, completeness, or suitability for any purpose ofthe
Content. Any opinions and views expressed in this publication are
the opinions and views of the authors,and are not the views of or
endorsed by Taylor & Francis. The accuracy of the Content
should not be reliedupon and should be independently verified with
primary sources of information. Taylor and Francis shall not
beliable for any losses, actions, claims, proceedings, demands,
costs, expenses, damages, and other liabilitieswhatsoever or
howsoever caused arising directly or indirectly in connection with,
in relation to or arising out ofthe use of the Content.
This article may be used for research, teaching, and private
study purposes. Any substantial or systematicreproduction,
redistribution, reselling, loan, sub-licensing, systematic supply,
or distribution in anyform to anyone is expressly forbidden. Terms
& Conditions of access and use can be found at
http://www.tandfonline.com/page/terms-and-conditions
-
GeneralStatistics and Ethics: Some Advice for Young
Statisticians
Stephen B. VARDEMAN and Max D. MORRIS
We write to young statisticians about the nature of statistics
andtheir responsibilitiesasmembers of the
statisticalprofession.Weobserve that the practiceof the discipline
is inherentlymoral andthat this fact has serious implications for
their work. In light ofthis, we offer some advice about how they
should resolve tothink and act.
KEY WORDS: Graduate study; Integrity; Principle; Profes-sional
practice; Research; Teaching.
Dear Gentle Reader:
So, you are embarking on a career in statistics. Good. It is
agenuinely noble pursuit, though this may be hard to see as
youwrestle with new-to-you technical issues varying from Howdo I
get this SAS job to run? to How do I show this thing isUMVU? and on
occasion nd yourselfwondering What is thepoint of all this?This
last question about purpose is actually a very important
and quite serious one. It has implications that run far
beyondyour present pain (and joy) of getting started. How you
an-swer it will affect not only you, but also the profession,
andhuman society at large. We write to offer some advice and
en-couragement, and to say how we hope you frame your answerto this
simultaneously practical and cosmic question.What are this subject
and this profession really all about?
And why are you doing what you are doing? For sure, thereare
details to learn (and keep current on throughout a career).There is
everything from the seemingly uncountable number oftricks of rst
year probability theory, to statistical computing,to nonlinear
models. It initially looks like soup to nuts. Youknow that
statistics is about collectingand handlingdata. That istrue, but
incomplete; there is much more than that at work here.The vital
point is that this discipline provides tools, patterns
of thought, and habits of heart that will allow you to deal
withdata with integrity. At its core statistics is not about
cleverness
Stephen B. Vardeman is Professor, and Max D. Morris is
Professor, De-partment of Statistics and Department of Industrial
and Manufacturing Sys-tems Engineering, Iowa State University Ames,
IA 50011-1210(E-mail: [email protected]). The authors
gratefully acknowledge the generous input of anumber of colleagues.
Karen Kafadar, Bob Stephenson, Bill Meeker, and DeanIsaacson
provideddetailed comments on a rst draft of this article. And the
inputof Ken Ryan, Tammy Brown, Mike Moon, Bill Notz, Tom Dubinin,
Frank Pe-ters, David Moore, Bill Duckworth, Bruce Held, Dennis
Gilliland, Bobby Mee,Doug Bonett, Dan Nettleton, and Hal Stern is
also gratefully acknowledged.
and technique, but rather about honesty. Its real contributionto
society is primarily moral, not technical. It is about doingthe
right thingwhen interpreting empirical information. Statis-ticians
are not the worlds best computer scientists, mathemati-cians, or
scienti c subjectmatter specialists.We are (potentially,at least)
the best at the principledcollection,summarization, andanalysis of
data. Our subject provides a framework for dealingtransparently and
consistently with empirical information fromall elds; means of
seeing and portraying what is true; ways ofavoiding being fooled by
both the ill intent (or ignorance) ofothers and our own incorrect
predispositions. The mix of the-ory and methods that you are
discovering is the best availablefor achieving these noble ends.
The more you practice with it,the sharper will become your
(fundamentallymoral) judgmentsabout what is appropriate in handling
empirical information.Others from areas ranging from philosophy to
physics might
well object that we have claimed too much, wrapping statisticsin
a cloak of virtue to the apparent exclusionof other
disciplines.After all, thoughtful scientists and humanists from a
variety of elds are engaged in the pursuit of truth. And any
serious ed-ucation has moral dimensions. Our point, however, is
that theparticular role that the profession plays in science and
societyshouldnot be viewed as amoral, and that this fact constrains
howwe all must think and act as its members.That society expects
our profession to play this kind of role
can be seen in the place statistics has as arbiter of what is
suf- cient evidence of ef cacy and safety to grant FDA approvalof a
drug, or enough evidence to support an advertisers claimfor the
effectiveness of a consumer product. And it can be seenin the fact
that many disciplines have statistical signi cancerequirements for
results appearing in their journals.Society also recognizes that
when statistical arguments are
abused, whether throughmalice or incompetence,genuine harmis
done. How else could a book titled How to Lie With Statis-tics
(Huff 1954) have ever been published and popular? Thefamous line
(attributed byMark Twain (1924) to BenjaminDis-raeli) There are
three kinds of lies: lies, damned lies, and statis-tics witnesses
effectively to societys distaste for obfuscationor outright
dishonesty cloaked in the garb of statistical technol-ogy. Society
disdains hypocrisy. It hates crooked lawyers, shadycorporate
executives, and corrupt accountants, and it has con-tempt for
statisticiansand statisticalwork that lack integrity.Butyoung
statisticians sometimes nd themselves being encour-aged to offer
questionable interpretations of data. This pres-sure can come even
from well-meaning individualswho believethat their only interest is
in ensuring that their position is treatedfairly. Maintaining an
independent and principled point-of-
c 2003 American Statistical Association DOI:
10.1198/0003130031072 The American Statistician, February 2003,
Vol. 57, No. 1 21
Dow
nloa
ded
by [T
IB &
Univ
ersita
etsbib
liothe
k] at
08:49
03 Ja
nuary
2015
-
view in such contexts is critical if a statistician hopes to
avoidbecoming a part of Disraelis third lie.So, you are embarking
upon a noble and serious business.We
take as given that you have a basic moral sense and a
strongdesire to personally do good. We also take as self-evident
thatintegrity is a pattern of life, not an incident. Principled
peopleconsistently do principledwork, regardless of whether it
servestheir short-termpersonal interests. Integrity is not
somethingthatis turned on and off at ones convenience. It cannot be
generallylacking and yet be counted on to appear in the nick of
timewhenthe greater good calls. This implies that what you choose
tothink and do now, early in your career, are very good
predictorsof what you will think and do throughout the whole of it.
Youare setting patterns that will endure over a professional
lifetimeand substantially in uence the nature and value of what you
canhope to accomplish.A fair amount has been written about
professional ethics in
statistics and we do not propose to review it all or comment
onevery issue that has been raised. For example, Demings
(1986)article is fundamentally a discussion of ethics. Both the
Amer-ican Statistical Association (1999) and the International
Statis-tical Institute (1985) have of cial statements on ethical
guide-lines for statisticians.And in a more general setting,
theNationalAcademy of Sciences (1995) has published a useful
booklet thatis primarily about ethics in science and has
implications for sta-tistical practice.Our more speci c goal here
is to suggest some things that
a high view of the discipline means for your present work
andattitudes. Aiming to speak to both statistics graduate
studentsand recent grads, well begin with some implications for
life ingraduate school, and then move on to implications for an
earlycareer in the discipline.
ADVICE FOR STATISTICS GRADUATE STUDENTS
Graduate student ethics (or for that matter professionalethics)
is really just plain ethics expressed in a graduate stu-dent (or
professional)world. A discussion of it really boils downto
consideration of circumstances and issues that arise in a
par-ticular graduate student (or professional) setting. So an
obviousplace to begin is with general student responsibilities. If
you arestill in graduate school, we urge you to be scrupulous about
yourconduct in the courses you take. Here are some speci cs:
Resolve to never accept credit for work that is not yourown. It
should make no difference to you whether an exam isproctored or
unproctored.Whatever the homework policy of thecourse, make it your
practice to clearly note on your papersplaceswhere you have gained
from discussionswith classmatesor consultingold problem sets of
others. Its simply right to giveothers credit where it is deserved
and its simply wrong to takecredit where it is undeserved.
If course policy is that everyone is completely on theirown,
resolve in advance to politely refuse to discuss with peerstopics
that are off-limits, even if others violate the policy. It mayseem
a small thing at the time, but you are setting life
trajectoriesthat are bigger than the particular incidents.
Determine to never take advantage of (or over) your peers.If you
join a group study session, be ready to make your faircontribution,
not just to bene t from the input of others. If youhave legitimate
access to old les or notes or textbooks that arehelpful, let others
know about them so that they can bene t aswell.
What do these three points say? Simply that you should playby
the rules set out and be clear and honest about all contri-butions
made to the work you turn in. Why would anyone dootherwise?
Honestly, only to gain an undeserved advantage in acourse grade, or
to avoid some effort. But a studentwilling to cutcorners for an A
or a free weekend will have serious dif cultynot cutting corners in
later professional responsibilities whenthe reward is a promotion
or pay raise or a free weekend.Some additional issues are related
to the notion of doing the
hard thing. Everyone has things that come harder for them
thanothers. Its human nature to want to avoid what is dif cult
andto even convince ourselves that really, the easy thing is what
isimportant and the hard thing is worthless. But that is not
onlyobviously silly, it has moral implications. Here is some
advicefor the student reader:
Understand that acquiring an advanced education is a dif- cult
enterprise, that there may be times when you feel likecomplaining
about this, but that it doesnt really help to do so.Whining wastes
energy and can poison the learning atmospherefor others.You are
engagedin a noble, if dif cult, pursuit.Give ityour best shot
without complaining.After all, most thingsworthdoing are hard.
Resolve to work on your weaknesses rather than excusethem.Doing
good statisticalwork is important, and demands thebest possible
personal tool kit. The reasoning I nd methods(theory) easier than
theory (methods), so Ill just do methods(theory) implicitly and
quite wrongly assumes that one can dogood statistical work with
half a tool kit.
Decide not to denigrate the strengths of others. Give
otherpeople credit for what they can do that you cannot. Find
yourniche without minimizing the honest efforts and contributionsof
others.
Determine to take the courses that will enable you to be
thebest-educated and most effective statistician you can be.
Theseare often academically demanding, and may not form a
par-ticularly easy route to a high GPA. While dif culty, per se,
isnot necessarily a measure of how often you will nd the ma-terial
in a course useful, it is related to the mental disciplineyou will
develop. If you choose a course that covers materialyou could
easily pick up on your own or because it is taughtby a professor
who demands little in exchange for an A, youvecheated yourself. The
choices you make about curriculum aremoral choices, not just
choices of convenience.You have a lim-ited time in graduate school
: : : use it wisely. How effective youwill be as a professional
depends on it. Besides, your choicessay something nontrivial about
the personal character that youare developing.
Purpose to do what your thesis or dissertation advisor setsfor
you to do, as independently as you can. While it may seemthat some
assignments are arbitrary or unnecessary, remember
22 General
Dow
nloa
ded
by [T
IB &
Univ
ersita
etsbib
liothe
k] at
08:49
03 Ja
nuary
2015
-
that you do not have your advisors experience as a researcher
oreducator.This person knowswhat you know, what your abilitiesare,
and the dif culty of your problem. He or she is trying tohelp you
to develop as a responsible and independent memberof the
profession, one accustomed to consistentlyworking up toyour
capabilities. Focusing your energy on the challenge of theproblem
and the opportunity it represents will take you muchfarther than
wasting your energy in grumbling or in negotiatingto be led through
every detail of a solution.
It is worth adding a further note related to this last point.
Theadvisoradvisee experience has the potential to be invigorat-ing
and rewarding (both professionally and personally) for bothparties.
Think of the efforts you put into it not only as a require-ment for
the degree, but as the beginning of what may be oneof your most
important and cherished long-term relationships.Find someone to
work with who you like and respect, and putyour energy into the
enterprise.Most statistics graduate studentswork as graduate
assistants.
Assistants should remember rst that an assistantship is not
afellowship, but rather a job. And it is axiomatic that
principledpeople return honest effort for their pay. If you are
working on afaculty members grant, that person must produce quality
workin line with the interests of some outside entity. Do what
youcan to help him or her. If you are a teaching assistant,
thereare lectures to conscientiously prepare and deliver, papers
tocarefully grade, and students to help. If you are a
consultant,people with real problems of data analysis will appear
at yourdoor seeking aid. They need your best effort and advice. Let
usamplify a bit:
If you are a research assistant it is understood that you
haveyour own class work and thesis or dissertation to attend to.But
some of your weekly hours are rst committed to providingthe help
(programming, library work, report writing, etc.) youremployer
needs. There are important educational bene ts thataccrue as you
practice at these duties. But the most fundamentalreason to carry
themout conscientiouslyand cheerfully is simplythat it is the right
thing to do. (And it is wrong to think that cuttingcorners now
doesnt say anythingabout later behavior. Life willalways be hectic
and there is no reason to expect your workhabits after nishing
school to be better than the ones you aredeveloping now.)
If you are a teaching assistant, purpose to make the bestof the
fact that along with some conscientious, motivated, andpleasant
students, youwill deal with some unpleasant, intention-ally
ignorant, lazy, and dishonest students. It simply comeswiththe
territory. For your part, make it a point tomodel
integrityandpurpose for all of them.Do your best to convey thatwhat
you areteaching them really does matter and how they do it matters
aswell. Resolve that whatever your style/personality (from
an-imated to reserved) your body language will convey a
genuinewillingness to help. The job takes patienceplan on it.
Resolveto treat all of your students well, whether or not their
behaviorin any sense merits that. And it should go without saying
thatalthough you want to be pleasant and approachable,
propriety
and impartiality dictate that you are their instructor or TA,
nottheir pal.
If your assignment is to help with statistical consulting,you
are already wrestling (at a trainee level) with some of theserious
issues faced by one segment of our profession.Carefullyconsider and
handle these now, as you begin to see how thehuman element of
statistical consulting requires thoughtfuland principled
discipline. Youre going to have to argue withyourself in
conversations like:
What looks to me like the thing that should be done wouldtake
two hours to explain and several more hours of mytime to implement,
while this client would be happy withsomethingless appropriate that
I could explain in vemin-utes : : :
This client reallywants A to be true, but these data
lookinconclusive : : :
This looks pretty much OK except for that oddity overthere that
the client doesnt really want to discuss : : :
Graduate Student Reader, keep your eyes open during thisgraduate
student experience. Watch your faculty and emulatethe ones who take
seriously what they do. There are some nerole models in our
university statistics departments, excellentmembers of the
profession. Find them, and learn as much as youcan about what they
think and how they practice statistics.
ADVICE FOR YOUNG PROFESSIONALSTATISTICIANS
Many of the themes weve introduced in the context of grad-uate
study have their logical extensions to early professionallife. But
there are also other matters that weve not yet raised.We proceed to
discuss some of the less obvious extrapolationsand further ethical
issues faced by young statisticians, organiz-ing our advice around
the topics of (1) research/publication, (2)teaching, and (3)
professional practice.If you have nished a Ph.D., you have been
introduced to
the craft of research in statistical theory or methods. You
arein a position to help develop the professions supporting bodyof
knowledge and to contribute to our journals. Its importantto
consider the corresponding responsibilities. These are tiedclosely
to a proper view of the purpose of publication in statis-tics.
Published statistical research should provide reliable
andsubstantial new theory or methodology that has genuine
poten-tial to ultimately help statisticians in the practice of the
disci-pline. Statistical publication should not be treated as a
game. Itis, and should be treated as, a serious and moral business.
Hereare some points of advice issuing from this high view of
whatthe research and publication activity is all about:
Resolve that if you choose to submit work for publication,it
will be complete and represent your best effort. Submittingpa-pers
of little intrinsic value, half-done work, or work sliced intosmall
pieces sent to multiple venues is an abuse of an
importantcommunicationsystemand is not honorablescholarship.It is
notthe job of editors or referees to proofread or complete your
pa-pers, or to insist that you follow up on important issues that
youknow exist. See the Lets just send it off and let the
reviewers
The American Statistician, February 2003, Vol. 57, No. 1 23
Dow
nloa
ded
by [T
IB &
Univ
ersita
etsbib
liothe
k] at
08:49
03 Ja
nuary
2015
-
sort it out impulse for what it is, a temptation to off-load
yourwork to someone else. And the Ill just submit this
half-donething to an outlet that will print anything strategy does
nothingof real value for anyone. It wastes time and effort of those
in thereview system, and when successful it dilutes our
literature.This makes important work harder to nd, and in the end
callsinto question our very reason to exist as a profession.
Purpose that when asked to do the job of a referee, you willdo
it thoroughly, impartially, and in as timely a manner as possi-ble.
There is no obvious short-term payoff to doingwhat is righthere.
But the integrity and currency of the scienti c publicationprocess
depend on competent and principled referees taking thejob
seriously. Resolve never to do a shoddy/cursory review job,or worse
yet to let calculationsabout personalities (and personaladvantage)
govern how you judge a piece of work. Even thoughmany statistics
journals use a double-blind system, the pro-fession is small, and
you will nd it increasingly rare that youhave no idea who authored
a paper you receive for review. Soremember that the spirit of the
blind review policy is honorable,and that you have an obligation to
conduct your review in thisspirit evenwhen you cannot be
completelyblind.And dowhatyou can as an individual to help x the
widely recognized prob-lem that the reviewprocess in statistics is
presentlymuch slowerthan in many other disciplines.
Decide to routinely take the advice of editors and
refereesregarding papers that you submit for publication.Occasions
arerare where editors or referees have it all wrong or purposely
treatan author unfairly. Most often, the advice they offer is
construc-tive and when followed substantially improves an article.
Untilan editor signals clearly that he or she has no further
interestin a piece you have submitted, you should almost always
makegood faith efforts to revise your paper in accord with his or
heradvice. Serial journal-shopping for a venue that will publish
asubmission with essentially no revision may minimize the
totaleffort an author expends on a paper, but the practice wastes
theoverall energy of the profession and has a negative effect on
theoverall quality of what is published.
Determine to be scrupulous about giving credit where itis due.
If another has contributed substantially to the contentof a paper,
co-authorship is typically appropriate and should beoffered. (On
the other hand, never list a colleague as co-authorof a paper until
you have that persons explicit permission to doso.) And include
acknowledgments of others deserving thanksfor less extensive, but
real, help with an article.
Resolve to acknowledge priority and the derivative natureof your
work with due humility. If after the fact of publicationyou nd that
some of your results can be found in earlier work,immediately send
an acknowledgment to that effect to the jour-nal where your paper
appeared. In writing your papers in the rst place, we encourage you
to be forthright and helpful aboutwhat you know is already
publishedon your subject, delineatingcarefullywhat others have
already said andwhere your new con-tribution lies. (No one ever
really starts from scratch. Dontfall prey to the temptation to
leave unsaid what you know isalready known, thinking that to do so
strengthens your own po-sition.)And never
borrowpublished/copyrightedwords, even ofyour own
authorship,without acknowledgment.To do so is pla-
giarism and is completely unacceptable. (This caution extends,by
the way, to thesis and dissertation work, even if that work isnever
submitted to a journal for formal publication.)
A note related to this last point: Avoiding plagiarism placesan
extra burden on students whose writing skills are not
strong,especially those struggling with English as a second
language.But it is essential to nd ones own words and not simply
copyor even paraphrase those of another (even for parts of a
paperthat are background and obviouslydont purport to provide
newtechnical content). This is a very serious integrity issue.Next,
lets consider issues relevant to teaching of statistics
as a professional. There are reasons to do this whether or
notyou have plans for a career at a college or university.
Teach-ing/training is increasinglydone in house by
corporationsandconsultants, and it could be argued that most
professional pre-sentationsare essentially teaching efforts. The
logical extensionof the advice offered above to graduate teaching
assistants is, ofcourse, relevant here. But there is an important
extra dimensionto discuss, related to the freedom and
responsibility that a pro-fessional has in answering the question
What will govern whatand how I teach?Will it be Whats easy for me?
Or will it beWhat will get the best short-term reaction from the
students?Or will it be My best professional judgementas to what the
stu-dents need for the long term and my best understanding of howto
effectively convey that information? This is a moral choice.Here is
some ampli cation:
Determine that you wont fall into the trap of organizingall
courses around your technical specialty. This is an issue
offundamental humility and recognition that none of us has put
allthat is needed into our personal little package (to say
nothingabout the matter of truth in advertising!). But we suspect
thatyou know what we are talking about, having seen people
turnevery course they teach into a platform to show off their
ownwork.
Purpose not to be governed by what is easy to do. This isnot an
entirely separate issue from the previous one. But we arealso
thinking about cases where the case is not so blatant or nottied
directly to ones specialty. Its a lot of work to learn newmethods
and software to include in a course, to freshen exam-ples, to
develop new laboratories and assignments for students,to replace
outdated topics and means of presentation. And itssometimes
possible to get bywithout investing that effort. Butdoing so is
simply wrong. We urge you not to take that route.
Resolve to do the best for your students,whether or not
theyappreciateyourefforts in the short term.We live in a
consumersociety.There is hugepressure on teachers in all
contextstomakestudents happy.But statistics is hard, and
studentsDONT knowwhat they need.Youwill.We hope that you opt to do
your best toprovide that, not simply what will get the best crowd
reaction.Lots of jokes, little in the way of course demands, and
highgrades can please many audiences.And leave students ignorant.Of
course we should aim to be engaging in our presentation ofour
subject. But the point of teaching is to genuinely
improvesubjectmatter knowledge and the reasoning powers of
students.It is not to produce feel-good experiences for them. (In
thisregard, we were recently dismayed to see an Iowa
communitycollege president quoted in the Des Moines Register (2001)
as
24 General
Dow
nloa
ded
by [T
IB &
Univ
ersita
etsbib
liothe
k] at
08:49
03 Ja
nuary
2015
-
proudly saying We are really a service organization rst and
aneducational institution second. While that may in fact be true,it
is a terrible commentary on the state of the institution.)
Those of you beginningacademic careers will face enormousdemands
for early success. Most universities require
substantialaccomplishments in both research and teaching during the
rstsix years of employment, and some place the bar so high
thatseemingly superhuman effort is required. If numbers of
refereedpublicationsand instructorevaluationsare the keys to
success,can you afford to have real qualityas your primary goal? Is
thereenough time in six short years to accomplishall that is
required ifyou takeour advice seriously?These are real and hard
questions.How you use your assistant professorship is critical to
your long-term professional success, and it is obvious that you
must takeyour institutions expectations into account. But, we urge
youas you face these issues to remember that one who spends
anassistant professorship cutting corners is at best prepared to
bean associate professorwho knows how to cut corners : : : not
onewho has learned how to make a difference.Turning nally to the
area of professional practice, we note
that most of what has been written about ethical guidelines
forstatisticians concerns what is appropriate in public practice,
inlending aid to others in the impartial and ef cient
collectionandanalysis of their data. This is understandable, as (1)
the disci-plineswhole reason to exist is ultimately to providesuch
aid and(2) this activity is both subtle and full of pitfalls. Both
the ethi-cal guidelines and public skepticism typi ed in the lies
quoteof Disraeli point to the fact that statistics can be used to
formhighly technical and even technically correct support for
state-ments which are in fact not true. We might hope this could
hap-pen onlywhen nonstatisticianspractice statistics without
propertechnical understandingof the subject. But statistical lies
are byde nition immoral uses of statistical arguments,whether
techni-cally correct or not, and stem from societal pressures that
affectstatisticians and nonstatisticians alike. What then must you
doin society to preserve the disciplines (and your own)
integrity?First, recognize that a professional statistician should
never
behave like a courtroom lawyer. The practice of law is basedon
an adversarial model in which each lawyer represents an as-signed
point of viewthat which will yield the most positiveoutcome for his
or her client. While the use of lies and inten-tionally misleading
statements is prohibited in legal proceed-ings, legal strategy
certainly does involve the selective use ofevidence so as to
present the truth (or some part of it) in the lightmost favorable
to a particular point of view. But a key aspect ofthismodel of
litigationis that decisions are made by an unbiasedauthority (a
judge or jury) based not on the case presented by asingle side, but
only after arguments presented by all parties
areheard.Statisticiansusuallydo not operate in
suchwell-controlledad-
versarial systems. If you do work in this kind of arena you
mustkeep absolutely clear the distinction between an objective
ana-lyst and an advocate, and never purport to be (or think
yourself)the rst when you are the second. If you are employed by an
or-ganization(whether on a permanent basis or as a
consultant)youare by de nition not disinterested in its well-being.
And if youare working pro bono for a cause you support, you are not
dis-
interested in furthering the cause. In either case, it is
axiomaticthat your professional judgment is potentially clouded by
whatyou (quite naturally) want to be true. And you will be no
fairjudge of the extent to which this clouding has occurred.
Thereis real danger here. There is little that is more damning to
thediscipline than for one of its professionals, implicitly
claimingsome degree of objectivity,to be publiclyexposed as
overstatinga statistical case in favor of his or her employer or
cause.More commonly, statisticians function as consultants to
those
who must make decisions. We do this through careful
andthoughtful design of data collection mechanisms and analysisof
assembled data. But careful and thoughtful here are wordsthat
acknowledge a critical fact: Statistical analysis of data canonly
be performed within the context of selected assumptions,models,
and/or prior distributions. A statistical analysis is ac-tually the
extraction of substantive information from data andassumptions. And
herein lies the rub, understood well by Dis-raeli and others
skepticalof ourwork: For givendata, an analysiscan usually be
selected which will result in informationmorefavorable to the owner
of the analysis than is objectively war-ranted.The only cure for
this dif culty is statistical practice based
on assumptions embodying an informed, balanced, and
honestrepresentation of what is known. Known, not wished
for,desired, convenient, or even other-than-worst-fears. Thishas
implications for how statisticiansmust be and act if they areto be
both effective and ethical.
Statisticians must be knowledgeable about the system un-der
study. They should not present themselves as competent toanalyze
data from systems aboutwhich they have no substantiveunderstanding.
Real data are not context-free.
On the other hand, statisticians must recognize and ac-knowledge
the limitations of their subject matter knowledge.Data and
variation are ubiquitous.Knowing how to handle themcan give you
important and even uncommon insights in a varietyof contexts where
you have limited subject matter credentials.But the fact that you
can make contributions in league with ex-perts in a variety of elds
doesnt substitute for credentials inthose elds. The credibility of
the statistical profession dependsupon its members being scrupulous
about what they know andwhat they dont know. Never forget that you
are not the contextexpert.
Statisticiansmust go out of theirway to see that their analy-ses
allow interpretationsof the availabledatawhich are tenablebut not
popular in the statisticians organization. This does notmean be a
troublemaker, but it does mean that you shouldcarefully think
through how available data would be interpretedby those with all
possible rational points of view.
Statisticiansmust write complete reports stating the resultsof
their entire informed thought processesincludingwhat theyknow, what
they have assumed, what they have decided cannotbe assumed, andwhat
conclusionstenableassumptions support.Our reports should contain
complete and suf cient analysesupon which any rational point of
view can be argued. If youcome to the conclusion that one of the
spectrum of sensibleinterpretations is best in a particular
application,make it yourgoal to be absolutely transparent about
your reasoning. People
The American Statistician, February 2003, Vol. 57, No. 1 25
Dow
nloa
ded
by [T
IB &
Univ
ersita
etsbib
liothe
k] at
08:49
03 Ja
nuary
2015
-
should be able to easily see your full set of model
assumptions,understandwhat methodologyyouhaveused tomake
inferencesin thatmodel, and have access to diagnosticand
robustnessworkyou have done. (This advice is sound in general. But
it is perhapsespecially relevant to explicitlyBayesian analyses. A
consumerof a posteriordistributionhas amoral right to knowhow
stronglyit depends upon the prior.) Honest statistical work has
nothingto hide. It says what it says. It doesnt try to obscure
pointswhere alternative conclusions are possible if other
assumptionsare made or different analysis paths are followed, and
admitswheremodel ts are short of perfectionor conclusionsare
highlymodel-dependent.
As a statistician,yourallegiancemustbe to nding the conclu-sions
which can be supported by data and careful assumptions.Does this
make the business of assumption selectionmore dif -cult than it
seemed in yourstatisticscoursework?Does it seem asthoughyoumust
take these issuesmore personallyand seriouslythanour favorite
semi-academicphrase LetX1; X2; : : : ;Xn beiidF : : : ? Does it
sound likeyour formulationof these assump-tions may have more to do
with nonmathematical values thanhas been discussed in your
textbooks?Yes, this andmore is true.Ethical statistical practice
requires that you take responsibilityfor acquiring substantive
understanding, knowing all rationalpoints of view, and making
decisions well beyond those basedentirely in data.
You must examine yourself to see that you are not
evensubconsciously leaning toward analyses which you believe
willplease the boss or yourself, or simplify the problem unjusti-
ably. This means that you cannot afford to think of yourselfas a
data technician or a hired gun. You must be secure enoughto
simultaneously separate any prior vested interest (yours orothers)
in the outcome from your analysis, and meld togetherseamlessly
everythingyou knowabout the subjectmatter of yourinvestigationwith
the structure of your statisticalwork. You can-not do this unless
you have strength of character and integrity.
You must not stop with the obvious or even the most
likelyexplanation of data, but nd ways to examine them so that
allrational viewpoints can be informed. This means that you
willwork harder and longer than anyonewho reads your reports
willever know. You will not rest until you know you understandall
the information contained in the data, where information
is de ned by the context of your work across the spectrum
ofrational viewpoints. You cannot do this unless you develop
anethic of self-reliance, thoroughness, and hard work.
You must understand fully what your assumptions say andwhat they
imply. You must not claim that the usual assump-tions are
acceptable due to the robustness of your techniqueunless you really
understand the implications and limits of thisassertion in the
context of your application.And youmust abso-lutely never use any
statisticalmethodwithout realizing that youare implicitlymaking
assumptions, and that the validity of yourresults can never be
greater than that of themost questionableofthese. You cannot do
this unless you remain dedicated to beingthe best
technicalstatisticianyou canpossiblybe, understandingthat this
involves knowing and understanding the mathematicalarguments aswell
as the computationaltechniquesbehindeverytool you need.
Well there it is, more than enough advice to keep a
youngstatistician busy for a career. We hope we dont sound too
muchlike myopic cranks, nding serious ethical issues to raise
ineven the most mundane contexts. Instead, we hope that we
haveargued effectively that ethical matters are central to our
disci-pline and provided some insight into issues that this raises.
Wefurther hope that you determine to take the matter of
principlemost seriously.
Carry on, Gentle Reader.
[Received April 2002. Revised November 2002.]
REFERENCES
American StatisticalAssociation (1999),EthicalGuidelines
forStatistical Prac-tice,
http://www.amstat.org/profession/ethicalstatistics.html
Deming, W. E. (1986), Principles of Professional Statistical
Practice, in En-cyclopedia of Statistical Science (vol. 7), eds. S.
Kotz and N. Johnson, NewYork: Wiley.
DesMoinesRegister (2001),Western IowaTech AmongNationsFastest
Grow-ing Schools, December 31, 2001, pp. B-1.
Huff, D. (1954),How to Lie with Statistics, New York:
Norton.International Statistical Institute (1985), Declaration on
Professional Ethics,http://www.cbs.nl/isi/ethics.htm.
National Academy of Sciences (1995),On Being a Scientist:
Responsible Con-duct in Research, Washington, DC: National Academy
Press. Also availableonline at
http://www.nap.edu/books/0309051967/html/.
Twain,M. (1924),Mark Twains Autobiography, New York: Harper
& Brothers.
26 General
Dow
nloa
ded
by [T
IB &
Univ
ersita
etsbib
liothe
k] at
08:49
03 Ja
nuary
2015
