Statistics and Ethics: Some Advice for Young Statisticians Author(s): Stephen B. Vardeman and Max D. Morris Source: The American Statistician, Vol. 57, No. 1 (Feb., 2003), pp. 21-26 Published by: American Statistical Association Stable URL: http://www.jstor.org/stable/3087273 Accessed: 22/04/2009 07:58
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=astata. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

American Statistical Association is collaborating with JSTOR to digitize, preserve and extend access to The American Statistician.

http://www.jstor.org

Gieneral

Statistics and Ethics: Some Advice for Young Statisticians and Stephen B. VARDEMAN Max D. MORRIS and technique, but ratherabout honesty. Its real contribution to society is primarilymoral, not technical. It is about doing the right thing when interpreting empiricalinformation.Statisticians are not the world's best computerscientists, mathematicians, or scientificsubjectmatterspecialists.Weare (potentially, at least) the best at theprincipledcollection, summarization, and of data. Our subjectprovides a frameworkfor dealing analysis and transparently consistentlywith empiricalinformationfrom all fields; means of seeing and portrayingwhat is true;ways of avoiding being fooled by both the ill intent (or ignorance) of others and our own incorrectpredispositions.The mix of theory and methods that you are discovering is the best available for achieving these noble ends. The more you practicewith it, will become your (fundamentally the sharper moral)judgments aboutwhat is appropriate handlingempiricalinformation. in Othersfrom areasrangingfrom philosophyto physics might well object that we have claimed too much, wrappingstatistics in a cloak of virtueto the apparent exclusionof otherdisciplines. After all, thoughtfulscientists and humanistsfrom a varietyof fields are engaged in the pursuitof truth.And any serious education has moral dimensions. Our point, however, is that the role that the profession plays in science and society particular shouldnot be viewed as amoral,andthatthis fact constrainshow we all must thinkand act as its members. That society expects our profession to play this kind of role can be seen in the place statisticshas as arbiterof what is sufficient evidence of efficacy and safety to grant FDA approval of a drug, or enough evidence to supportan advertiser'sclaim for the effectiveness of a consumerproduct.And it can be seen in the fact that many disciplines have "statisticalsignificance" for requirements resultsappearingin theirjournals. also recognizes that when statistical argumentsare Society abused,whetherthroughmalice or incompetence,genuineharm is done. How else could a book titled How to Lie WithStatistics (Huff 1954) have ever been published and popular?The famousline (attributed MarkTwain(1924) to BenjaminDisby raeli) "Therearethreekindsof lies: lies, damnedlies, and statistics" witnesses effectively to society's distaste for obfuscation or outrightdishonestycloaked in the garbof statisticaltechnology. Society disdainshypocrisy.It hatescrookedlawyers,shady corporateexecutives, and corruptaccountants,and it has contemptfor statisticiansandstatisticalworkthatlack integrity.But young statisticianssometimes find themselves being "encourof aged" to offer questionableinterpretations data. This pressurecan come even from well-meaningindividualswho believe thattheironly interestis in ensuringthattheirposition is treated "fairly."Maintainingan independentand principledpoint-ofThe American Statistician, February 2003, Vol. 57, No. 1 21

We write to young statisticiansaboutthe natureof statisticsand theirresponsibilitiesas membersof the statisticalprofession.We observethatthe practiceof the disciplineis inherentlymoraland that this fact has serious implicationsfor their work. In light of this, we offer some advice about how they should resolve to think and act. KEY WORDS: Graduatestudy; Integrity;Principle;Professional practice;Research;Teaching.

Dear Gentle Reader: So, you are embarkingon a careerin statistics.Good. It is a genuinely noble pursuit,though this may be hardto see as you wrestle with new-to-you technical issues varying from "How do I get this SAS job to run?"to "How do I show this thing is UMVU?"andon occasion findyourselfwondering"Whatis the point of all this?" This last question aboutpurposeis actuallya very important and quite serious one. It has implications that run far beyond How you anyour present pain (and joy) of "gettingstarted." swer it will affect not only you, but also the profession, and human society at large. We write to offer some advice and encouragement,and to say how we hope you frame your answer to this simultaneouslypracticaland cosmic question. What are this subject and this profession really all about? And why are you doing what you are doing? For sure, there are details to learn (and keep currenton throughouta career). Thereis everythingfrom the seemingly uncountablenumberof tricks of first year probabilitytheory,to statisticalcomputing, to nonlinearmodels. It initially looks like "soup to nuts."You know thatstatisticsis aboutcollecting andhandlingdata.Thatis true,but incomplete;thereis much more thanthatat work here. The vital point is that this discipline providestools, patterns of thought,and habits of heartthat will allow you to deal with data with integrity.At its core statisticsis not about cleverness
Stephen B. Vardemanis Professor, and Max D. Morris is Professor, Departmentof Statistics and Departmentof Industrialand ManufacturingSystems Engineering,Iowa State UniversityAmes, IA 50011-1210 (E-mail:vardeThe authorsgratefullyacknowledgethe generous inputof a man@iastate.edu). numberof colleagues. KarenKafadar, Bob Stephenson,Bill Meeker,and Dean Isaacsonprovideddetailedcommentson a firstdraftof this article.And the input of Ken Ryan, TammyBrown, Mike Moon, Bill Notz, Tom Dubinin,FrankPeters, David Moore, Bill Duckworth,Bruce Held, Dennis Gilliland,Bobby Mee, Doug Bonett, Dan Nettleton,and Hal Sternis also gratefullyacknowledged.

?) 2003 AmericanStatisticalAssociation DOI: 10.1198/0003130031072

view in such contexts is critical if a statisticianhopes to avoid becoming a partof Disraeli's third"lie." So, you areembarking upona noble and seriousbusiness.We take as given that you have a basic moral sense and a strong desire to personallydo good. We also take as self-evident that integrityis a patternof life, not an incident. Principledpeople consistentlydo principledwork,regardlessof whetherit serves is theirshort-term personalinterests.Integrity not somethingthat is turnedon and off at one's convenience.It cannotbe generally lackingandyet be countedon to appearin the nick of time when the greater good calls. This implies that what you choose to thinkand do now, early in your career,are very good predictors of what you will think and do throughoutthe whole of it. You are settingpatternsthatwill endureover a professionallifetime influencethe natureandvalue of whatyou can and substantially to accomplish. hope A fair amounthas been writtenabout professionalethics in statisticsand we do not proposeto review it all or commenton every issue thathas been raised.For example,Deming's (1986) articleis fundamentallya discussion of ethics. Both the AmerStatisican StatisticalAssociation (1999) and the International tical Institute(1985) have official statementson ethical guideAnd in a moregeneralsetting,the National lines for statisticians. Academyof Sciences (1995) has publisheda useful bookletthat is primarilyaboutethics in science andhas implicationsfor statistical practice. Our more specific goal here is to suggest some things that a high view of the discipline means for your presentwork and attitudes.Aiming to speak to both statistics graduatestudents and recentgrads,we'll begin with some implicationsfor life in graduateschool, and then move on to implicationsfor an early careerin the discipline. ADVICE FOR STATISTICS GRADUATE STUDENTS "Graduatestudent ethics" (or for that matter "professional ethics")is reallyjust "plainethics"expressedin a graduatestudent(or professional)world.A discussionof it reallyboils down to considerationof circumstancesand issues that arise in a particulargraduatestudent(or professional)setting. So an obvious place to begin is with generalstudentresponsibilities.If you are still in graduate school, we urgeyou to be scrupulousaboutyour conductin the courses you take. Here are some specifics: * Resolve to never accept credit for work that is not your own. It should make no difference to you whetheran exam is Whateverthe homeworkpolicy of the or proctored unproctored. course, make it your practice to clearly note on your papers places whereyou have gainedfrom discussionswith classmates or consultingold problemsets of others.It's simply rightto give otherscreditwhere it is deservedand it's simply wrong to take creditwhere it is undeserved. * If course policy is that everyone is "completely on their own,"resolve in advanceto politely refuse to discuss with peers topics thatareoff-limits, even if othersviolate the policy. It may seem a smallthingat the time, butyou aresettinglife trajectories incidents. thatare bigger thanthe particular
22 General

* Determineto nevertake advantageof (or over) your peers. If you join a group study session, be ready to make your fair not contribution, just to benefit from the input of others.If you have legitimateaccess to old files or notes or textbooksthatare helpful, let others know about them so that they can benefit as well. What do these threepoints say? Simply thatyou shouldplay by the rules set out and be clear and honest about all contributions made to the work you turn in. Why would anyone do otherwise?Honestly,only to gain an undeservedadvantagein a coursegrade,or to avoid some effort.But a studentwilling to cut corers for an A or a free weekend will have serious difficulty not cutting corners in later professional responsibilitieswhen the rewardis a promotionor pay raise or a free weekend. Some additionalissues arerelatedto the notion of "doingthe hardthing."Everyonehas thingsthatcome harderfor themthan others. It's humannatureto want to avoid what is difficultand to even convince ourselves thatreally, the easy thing is what is importantand the hardthing is worthless. But that is not only obviously silly, it has moral implications.Here is some advice for the studentreader: * Understandthat acquiringan advancededucationis a difficult enterprise,that there may be times when you feel like complainingabout this, but that it doesn't really help to do so. Whiningwastes energyandcan poison the learningatmosphere for others.Youareengagedin a noble, if difficult,pursuit.Give it yourbest shot withoutcomplaining.Afterall, most thingsworth doing are hard. * Resolve to work on your weaknesses ratherthan excuse them.Doing good statisticalworkis important, demandsthe and best possible personaltool kit. The reasoning "I find methods (theory) easier than theory (methods), so I'll just do methods (theory)"implicitly and quite wrongly assumes that one can do good statisticalwork with half a tool kit. * Decide not to denigratethe strengthsof others.Give other people credit for what they can do that you cannot. Find your niche without minimizing the honest efforts and contributions of others. * Determineto takethe coursesthatwill enable you to be the best-educatedand most effective statisticianyou can be. These are often academically demanding, and may not form a particularlyeasy route to a high GPA. While difficulty,per se, is not necessarily a measure of how often you will find the material in a course useful, it is related to the mental discipline you will develop. If you choose a course that covers material you could easily pick up on your own or because it is taught by a professorwho demandslittle in exchange for an A, you've cheated yourself. The choices you make about curriculumare moralchoices, notjust choices of convenience.You have a limited time in graduateschool ... use it wisely. How effective you will be as a professionaldepends on it. Besides, your choices say somethingnontrivialaboutthe personalcharacterthat you are developing. * Purposeto do what your thesis or dissertationadvisorsets for you to do, as independentlyas you can. While it may seem or that some assignmentsare arbitrary unnecessary,remember

thatyou do not haveyour advisor'sexperienceas a researcher or This personknows whatyou know,whatyourabilities educator. are, and the difficulty of your problem. He or she is trying to help you to develop as a responsibleand independentmember of the profession,one accustomedto consistentlyworkingup to your capabilities.Focusing your energy on the challenge of the problem and the opportunityit representswill take you much fartherthanwasting your energy in grumblingor in negotiating to be led throughevery detail of a solution. It is worthaddinga furthernote relatedto this last point. The advisor-advisee experience has the potential to be invigorating and rewarding(bothprofessionallyandpersonally)for both parties.Thinkof the effortsyou put into it not only as a requirement for the degree, but as the beginning of what may be one of your most importantand cherishedlong-termrelationships. Find someone to work with who you like and respect, and put your energy into the enterprise. Most statisticsgraduatestudentswork as graduateassistants. Assistants should rememberfirst that an assistantshipis not a fellowship, but rathera job. And it is axiomaticthat principled people returnhonest effortfor theirpay.If you areworkingon a faculty member'sgrant,thatperson must producequalitywork in line with the interests of some outside entity. Do what you can to help him or her. If you are a teaching assistant, there are lectures to conscientiously prepareand deliver, papers to carefully grade, and students to help. If you are a consultant, people with real problemsof data analysis will appearat your door seeking aid. They need your best effort and advice. Let us amplify a bit: * If you area researchassistantit is understood you have that own" class work and thesis or dissertationto attendto. "your But some of your weekly hoursare firstcommittedto providing the help (programming, librarywork, reportwriting,etc.) your needs. There are importanteducationalbenefits that employer accrueas you practiceat these duties. But the most fundamental reasonto carrythemout conscientiouslyandcheerfullyis simply thatit is the rightthingto do. (Andit is wrongto thinkthatcutting corers now doesn't say anythingaboutlaterbehavior.Life will always be hectic and there is no reason to expect your work habits after finishing school to be betterthan the ones you are developing now.) * If you are a teaching assistant,purpose to make the best of the fact that along with some conscientious, motivated,and intentionpleasantstudents,you will deal with some unpleasant, ally ignorant,lazy, and dishoneststudents.It simply comes with the territory. yourpart,makeit a pointto model integrityand For purposefor all of them. Do yourbest to convey thatwhatyou are teaching them really does matterand how they do it mattersas well. Resolve that whateveryour "style"/personality (from animated to reserved)your body language will convey a genuine willingness to help. Thejob takespatience-plan on it. Resolve to treatall of your studentswell, whetheror not their behavior in any sense merits that. And it should go without saying that althoughyou want to be pleasant and approachable, propriety

and impartialitydictate that you are their instructoror TA, not theirpal. * If your assignment is to help with statistical consulting, level) with some of the you are alreadywrestling(at a "trainee" seriousissues faced by one segmentof ourprofession.Carefully consider and handle these now, as you begin to see how the "humanelement" of statistical consulting requires thoughtful and principleddiscipline. You're going to have to argue with yourself in conversationslike: - Whatlooks to me like the thingthatshouldbe done would take two hours to explain and several more hours of my time to implement,while this client would be happywith that somethingless appropriate I could explainin five minutes ...

-

This client really wants "A" be true,but these datalook to inconclusive ...

- This looks pretty much OK except for that oddity over therethatthe client doesn't really want to discuss ... GraduateStudent Reader,keep your eyes open during this graduatestudent experience. Watch your faculty and emulate the ones who take seriously what they do. There are some fine role models in our university statistics departments,excellent membersof the profession.Findthem, andlearnas muchas you can aboutwhat they thinkand how they practicestatistics. ADVICE FOR YOUNG PROFESSIONAL STATISTICIANS Many of the themes we've introducedin the context of graduate study have their logical extensions to early professional life. But there are also other mattersthat we've not yet raised. We proceed to discuss some of the less obvious extrapolations and furtherethical issues faced by young statisticians,organizing our advice aroundthe topics of (1) research/publication, (2) teaching,and (3) professionalpractice. If you have finished a Ph.D., you have been introducedto the craft of researchin statisticaltheory or methods. You are in a position to help develop the profession's supportingbody of knowledge and to contributeto our journals. It's important to consider the correspondingresponsibilities.These are tied closely to a properview of the purposeof publicationin statistics. Published statisticalresearchshould provide reliable and substantialnew theoryor methodologythathas genuine potential to ultimatelyhelp statisticiansin the practice of the discipline. Statisticalpublicationshould not be treatedas a game. It is, and should be treatedas, a serious and moralbusiness. Here are some points of advice issuing from this high view of what the researchand publicationactivityis all about: * Resolve thatif you choose to submitwork for publication, it will be completeandrepresent yourbest effort.Submitting paof little intrinsicvalue, half-donework, or work sliced into pers small pieces sent to multiplevenues is an abuseof an important communication It systemandis not honorablescholarship. is not the job of editors or refereesto proofreador complete your paissues thatyou pers, or to insist thatyou follow up on important know exist. See the "Let'sjust send it off and let the reviewers
TheAmericanStatistician,February2003, Vol.57, No. 1 23

sort it out"impulse for what it is, a temptationto off-load your work to someone else. And the "I'll just submitthis half-done thing to an outletthatwill printanything" strategydoes nothing of real value for anyone.It wastes time and effortof those in the review system, and when "successful"it dilutes our literature. This makes importantwork harderto find, and in the end calls into questionour very reasonto exist as a profession. * Purposethatwhen askedto do thejob of a referee,you will do it thoroughly, and impartially, in as timely a manneras possible. Thereis no obvious short-term payoff to doing whatis right here. But the integrityand currencyof the scientificpublication process dependon competentandprincipledrefereestakingthe reviewjob, job seriously.Resolve neverto do a shoddy/cursory or worse yet to let calculationsaboutpersonalities(andpersonal governhow you judge a piece of work.Even though advantage) statisticsjournals use a "double-blind" system, the promany fession is small, and you will find it increasinglyrarethat you have no idea who authoreda paper you receive for review. So rememberthatthe spirit of the blindreview policy is honorable, and that you have an obligation to conduct your review in this And do what spiriteven when you cannotbe completely"blind." can as an individualto help fix the widely recognizedprobyou lem thatthe reviewprocessin statisticsis presentlymuch slower thanin many otherdisciplines. * Decide to routinelytake the advice of editors and referees regardingpapersthatyou submitfor publication.Occasionsare rarewhereeditorsor refereeshaveit all wrongor purposelytreat an authorunfairly.Most often, the advice they offer is constructive and when followed substantiallyimprovesan article.Until an editor signals clearly that he or she has no furtherinterest in a piece you have submitted,you should almost always make good faith efforts to revise your paperin accordwith his or her for advice. Serialjournal-shopping a venue that will publish a submissionwith essentially no revision may minimize the total effort an authorexpends on a paper,but the practicewastes the overallenergy of the professionand has a negativeeffect on the overall qualityof what is published.

giarismand is completely unacceptable.(This cautionextends, by the way, to thesis and dissertationwork, even if thatwork is never submittedto a journalfor formalpublication.) A note relatedto this last point: Avoiding plagiarismplaces an extraburdenon studentswhose writing skills are not strong, especially those strugglingwith English as a second language. But it is essential to find one's own words and not simply copy or even paraphrase those of another(even for parts of a paper thatarebackground obviously don't purport providenew and to technicalcontent).This is a very serious integrityissue. Next, let's consider issues relevantto teaching of statistics as a professional. There are reasons to do this whether or not you have plans for a career at a college or university.Teachis and ing/training increasinglydone "inhouse"by corporations and it could be arguedthat most professionalpreconsultants, sentationsareessentiallyteachingefforts.The logical extension of the advice offeredabove to graduateteachingassistantsis, of extradimension course, relevanthere. But thereis an important to discuss, relatedto the freedom and responsibilitythat a professional has in answeringthe question"Whatwill governwhat andhow I teach?"Will it be "What'seasy for me?"Or will it be "Whatwill get the best short-term reactionfrom the students?" Orwill it be "Mybest professional judgementas to whatthe students need for the long termand my best understanding how of to effectively convey thatinformation?" This is a moralchoice. Here is some amplification: * Determinethat you won't fall into the trap of organizing all courses aroundyour technical specialty.This is an issue of fundamental humilityandrecognitionthatnone of us has put all that is needed into our personal little package (to say nothing aboutthe matterof "truth advertising!"). we suspectthat in But you know what we are talking about, having seen people turn every course they teach into a platformto show off their own work.

* Purposenot to be governedby what is easy to do. This is not an entirelyseparateissue from the previousone. But we are also thinkingaboutcases wherethe case is not so blatantor not * Determine to be scrupulousabout giving credit where it tied directly to one's specialty. It's a lot of work to learn new is due. If anotherhas contributedsubstantiallyto the content methods and softwareto include in a course, to freshen examand and is of a paper,co-authorship typically appropriate should be ples, to develop new laboratories assignmentsfor students, offered. (On the otherhand,never list a colleague as co-author to replace outdatedtopics and means of presentation.And it's of a paperuntil you have thatperson'sexplicit permissionto do sometimespossible to "getby" withoutinvestingthateffort.But so.) And include acknowledgmentsof others deserving thanks doing so is simply wrong. We urge you not to take thatroute. for less extensive, but real, help with an article. * Resolve to do the best for yourstudents,whetheror not they * Resolve to acknowledgepriorityand the derivativenature appreciate youreffortsin the shortterm.We live in a "consumer" of your work with due humility.If after the fact of publication society.Thereis hugepressureon teachersin all contextsto make you find that some of your resultscan be found in earlierwork, studentshappy.But statisticsis hard,andstudentsDON'T know to immediatelysend an acknowledgment thateffect to the jour- whatthey need. Youwill. We hope thatyou opt to do yourbest to nal where your paper appeared.In writing your papers in the provide that, not simply what will get the best crowd reaction. and firstplace, we encourageyou to be forthright helpful about Lots of jokes, little in the way of course demands, and high know is alreadypublishedon yoursubject,delineating gradescan please manyaudiences.And leave studentsignorant. whatyou carefullywhatothershavealreadysaidandwhereyournew con- Of course we should aim to be engaging in our presentationof tributionlies. (No one ever really "startsfrom scratch."Don't our subject. But the point of teaching is to genuinely improve fall prey to the temptationto leave unsaid what you know is subjectmatterknowledgeandthe reasoningpowersof students. your own po- It is not to produce feel-good experiences for them. (In this alreadyknown,thinkingthatto do so strengthens And neverborrowpublished/copyrighted words,even of regard,we were recently dismayed to see an Iowa community sition.) To withoutacknowledgment. do so is pla- college presidentquoted in the Des Moines Register (2001) as your own authorship,
General

24

the firstandan interestedin furthering cause. In eithercase, it is axiomatic proudlysaying "Wearereally a service organization that your professionaljudgmentis potentiallyclouded by what educationalinstitutionsecond."While that may in fact be true, it is a terriblecommentaryon the state of the institution.) you (quite naturally)want to be true. And you will be no fair judge of the extent to which this clouding has occurred.There Those of you beginningacademiccareerswill face enormous is real dangerhere. There is little that is more damningto the demandsfor early success. Most universitiesrequiresubstantial discipline than for one of its professionals,implicitly claiming accomplishmentsin both researchand teaching duringthe first some degreeof objectivity,to be publiclyexposed as overstating six years of employment,and some place the bar so high that a statisticalcase in favorof his or her employeror cause. If effortis required. numbersof refereed seemingly superhuman Morecommonly,statisticiansfunctionas consultantsto those evaluationsarethe "keysto success," who must make decisions. We do this publicationsandinstructor through careful and can you affordto have real qualityas yourprimary goal? Is there thoughtfuldesign of data collection mechanisms and analysis if of assembleddata.But "carefuland thoughtful" enoughtime in six shortyearsto accomplishall thatis required here are words you takeouradvice seriously?These arerealandhardquestions. that acknowledgea criticalfact: Statisticalanalysis of data can is How you use yourassistantprofessorship criticalto yourlong- only be performedwithin the context of selected assumptions, term professional success, and it is obvious that you must take models, and/or prior distributions.A statisticalanalysis is acyour institution'sexpectationsinto account. But, we urge you tually the extractionof substantiveinformationfrom data and as you face these issues to rememberthat one who spends an assumptions.And herein lies the rub, understoodwell by Disassistantprofessorshipcutting corers is at best preparedto be raeliandothersskepticalof ourwork:Forgiven data,an analysis an associateprofessorwho knows how to cut comers ... not one can usuallybe selected which will resultin "information" more who has learnedhow to make a difference. favorableto the owner of the analysis than is objectively warTurningfinally to the area of professionalpractice,we note ranted. The only "cure" this difficultyis statisticalpracticebased for that most of what has been writtenabout ethical guidelines for on assumptionsembodying an informed,balanced,and honest in statisticiansconcerns what is appropriate public practice,in of lending aid to othersin the impartialandefficientcollection and representation what is known. "Known,"not "wished for," or This "convenient," even "other-than-worst-fears." as analysis of their data. This is understandable, (1) the disci- "desired," has implicationsfor how statisticiansmustbe and act if they are pline's whole reasonto exist is ultimatelyto providesuch aid and (2) this activity is both subtle and full of pitfalls. Both the ethi- to be both effective and ethical. cal guidelines and public skepticismtypifiedin the "lies"quote * Statisticiansmust be knowledgeableabout the system unof Disraeli point to the fact that statistics can be used to form der study. They should not presentthemselves as competentto highly technical and even technically correctsupportfor state- analyzedatafrom systems aboutwhich they haveno substantive ments which are in fact not true.We might hope this could hap- understanding. Real dataare not "context-free." practicestatisticswithoutproper pen only when nonstatisticians * On the other hand, statisticians must recognize and acof technicalunderstanding the subject.But statisticallies areby definitionimmoraluses of statisticalarguments, whethertechni- knowledgethe limitationsof their "subjectmatter"knowledge. Data andvariationareubiquitous.Knowinghow to handlethem cally corrector not, and stem from societal pressuresthataffect and statisticiansand nonstatisticiansalike. What then must you do can give you important even uncommoninsightsin a variety of contexts where you have limited subject mattercredentials. in society to preservethe discipline's (and your own) integrity? But the fact thatyou can make contributions league with exin First, recognize that a professional statistician should never in a variety of fields doesn't substitutefor credentialsin behave like a courtroomlawyer. The practice of law is based perts those fields. The credibilityof the statisticalprofessiondepends model in which each lawyer representsan ason an adversarial upon its membersbeing scrupulousaboutwhat they know and signed point of view-that which will yield the most positive what they don't know.Never forget thatyou are not the context outcome for his or her client. While the use of lies and intenexpert. tionally misleading statementsis prohibitedin legal proceed* Statisticiansmustgo out of theirway to see thattheiranalyings, legal strategycertainly does involve the selective use of evidence so as to presentthe truth(or some partof it) in the light ses allow interpretations the available data whichare tenable of most favorableto a particular of view. But a key aspect of but not popular in the statistician'sorganization.This does not point but this model of litigationis thatdecisions aremadeby an unbiased mean "be a troublemaker," it does mean that you should or jury) based not on the case presentedby a carefullythinkthroughhow availabledatawould be interpreted authority(a judge single side, but only afterarguments presentedby all partiesare by those with all possible rationalpoints of view. heard. * Statisticiansmustwritecompletereportsstatingthe results Statisticians usuallydo not operatein suchwell-controlledad- of theirentireinformedthoughtprocesses-including what they versarialsystems. If you do work in this kind of arenayou must know,what they have assumed,what they have decided cannot keep absolutely clear the distinctionbetween an objective ana- be assumed,and whatconclusionstenableassumptions support. to lyst and an advocate,and neverpurport be (or thinkyourself) Our reportsshould contain "completeand sufficient"analyses the firstwhen you arethe second. If you areemployedby an or- upon which any rationalpoint of view can be argued. If you basis or as a consultant)you come to the conclusion that one of the spectrumof sensible ganization(whetheron a permanent are by definitionnot disinterestedin its well-being. And if you interpretations "best"in a particular is application,makeit your areworking"probono"for a cause you support,you arenot dis- goal to be absolutelytransparent aboutyour reasoning.People
TheAmericanStatistician,February2003, Vol.57, No. 1 25

* You must understand fully what your assumptionssay and what they imply. You must not claim that the "usual assumptions" are acceptable due to the robustnessof your technique unless you really understand implicationsand limits of this the assertionin the contextof yourapplication.And you must absolutely neveruse any statisticalmethodwithoutrealizingthatyou areimplicitly makingassumptions,and thatthe validityof your resultscan neverbe greaterthanthatof the most questionableof these. Youcannot do this unless you remaindedicatedto being mustbe to findingthe conclu- thebest technicalstatisticianyou canpossibly be, understanding As a statistician, yourallegiance the sions which can be supportedby data and carefulassumptions. that this involvesknowingand understanding mathematical as Does this makethe businessof assumptionselection more diffi- arguments well as the computational techniquesbehindevery Does it seem as tool you need. cult thanit seemedin yourstatisticscoursework? thoughyou musttakethese issues morepersonallyandseriously Well there it is, more than enough advice to keep a young thanourfavoritesemi-academic phrase"LetX1, X2,..., X, be statisticianbusy for a career.We hope we don't soundtoo much of iid F ... ?"Does it soundlike yourformulation these assump- like myopic cranks,finding "seriousethical issues" to raise in values than even the most mundanecontexts. more to do with nonmathematical tions may have Instead,we hope thatwe have has been discussedin yourtextbooks?Yes, this andmoreis true. arguedeffectively that ethical mattersare centralto our disciEthical statisticalpracticerequiresthat you take responsibility pline and providedsome insight into issues thatthis raises. We for acquiring substantiveunderstanding, knowing all rational furtherhope that you determineto take the matterof principle points of view, and making decisions well beyond those based most seriously. entirelyin data. Carryon, Gentle Reader. * You must examine yourself to see that you are not even subconsciouslyleaning towardanalyseswhich you believe will [ReceivedApril 2002. RevisedNovember2002.] "please the boss" or yourself, or simplify the problemunjustifiably. This means that you cannot afford to think of yourself REFERENCES as a datatechnicianor a hired gun. You must be secure enough to simultaneouslyseparateany prior vested interest (yours or AmericanStatisticalAssociation(1999), "EthicalGuidelinesfor StatisticalPractice,"http://www.amstat.org/profession/ethicalstatistics.html others') in the outcome from your analysis, and meld together Deming, W. E. (1986), "Principlesof ProfessionalStatisticalPractice,"in Enseamlesslyeverythingyou know aboutthe subjectmatterof your cyclopedia of StatisticalScience (vol. 7), eds. S. Kotz and N. Johnson,New of York:Wiley. investigationwith the structure yourstatisticalwork. YoucanIowa not do this unless you have strengthof characterand integrity. Des MoinesRegister(2001), "Western TechAmongNation'sFastestGrow* You must not stop with the obvious or even the most likely explanationof data, but find ways to examine them so that all rationalviewpoints can be informed.This means that you will workharderandlongerthananyonewho readsyourreportswill ever know. You will not rest until you know you understand all the informationcontainedin the data, where "information" ing Schools,"December 31, 2001, pp. B-1. Huff, D. (1954), How to Lie with Statistics,New York:Norton. International StatisticalInstitute(1985), "Declarationon ProfessionalEthics," http://www.cbs.nl/isi/ethics.htm. NationalAcademy of Sciences (1995), On Being a Scientist:ResponsibleConduct in Research,Washington,DC: NationalAcademy Press. Also available online at http://www.nap.edu/books/0309051967/html/. New York:Harper& Brothers. Twain,M. (1924), MarkTwain'sAutobiography,

shouldbe able to easily see your full set of model assumptions, understand whatmethodologyyou haveused to makeinferences in thatmodel, andhaveaccess to diagnosticandrobustnesswork you havedone. (This adviceis soundin general.But it is perhaps especially relevantto explicitly Bayesian analyses.A consumer of a posteriordistribution a moralrightto knowhow strongly has it dependsupon the prior.)Honest statisticalwork has nothing to hide. It says what it says. It doesn't try to obscure points where alternative conclusions are possible if otherassumptions are made or different analysis paths are followed, and admits wheremodel fits areshortof perfectionorconclusionsarehighly model-dependent.

is defined by the context of your work across the spectrumof rationalviewpoints. Youcannot do this unless you develop an ethic of self-reliance,thoroughness,and hardwork.

26

General