American Journal of Epidemiology Vol. 133, No 5Copyright C 1991 by The Johns Hopkins Urtverstty School of Hygiene and Public Health Pnnted in U.S.A.All rights reserved

REVIEWS AND COMMENTARY

When Genius Errs: R. A. Fisher and the Lung CancerControversy

Paul D. Stolley

R. A. Fisher's work on lung cancer and smoking is critically reviewed. The controversyis placed in the context of his career and personality. Although Fisher made invaluablecontributions to the field of statistics, his analysis of the causal association betweenlung cancer and smoking was flawed by an unwillingness to examine the entire bodyof data available and prematurely drawn conclusions. His views may also have beeninfluenced by personal and professional conflicts, by his work as a consultant to thetobacco industry, and by the fact that he was himself a smoker. Am J Epidemiol 1991;133:416-25.

epidemiology; history of medicine, 20th cent.; lung neoplasms; smoking; statistics

Editor's note: For a discussion of this pa-per and for the author's response, see pages426 and 428, respectively, and page 429.

AU men are liable to error, and most men are, inmany points, by passion or interest, under temp-tation to i t . . .

—John Locke, An Essay Concerning HumanUnderstanding, Chap 20, Sec 17

In an article published in 1989, JanVandenbroucke (1) briefly reviewed the lungcancer controversy three decades ago whentwo prominent statisticians, J. Berkson andR. A. Fisher, doubted that cigarettes causedlung cancer. Vandenbroucke described the".. .well-written and cogent papers thatmight have become textbook classics for

Received for publication May 15, 1990, and in finalform August 13,1990.

From the Clinical Epidemiology Unit, University ofPennsylvania School of Medicine, 220-L Nursing Edu-cation Building, Philadelphia, PA 19104-6095. (Reprintrequests to Dr. Paul 0. Stolley at this address.)

The author would ike to thank Professor RichardDoll for reviewing the manuscript and Stephanie C.Austin and Myma G. Bolt for helping to revise it.

their impeccable logic and clear expositionof data and argument if only the authorshad been on the right side..." (1, p. 3).

My reading of everything Fisher everpublished on the subject—five letters, twolectures, and an article—revealed almostprecisely the opposite of Vandenbroucke'sevaluation: incomplete and highly selecteddata (or no data but much speculation), withscant attempts to weigh the evidence or re-veal the obvious deficiencies in his data.Fisher sounds like a man with "an axe togrind." I have tried to piece together thestory of his involvement in this controversyand to place it in the context of his careerand personality. Fisher was not just on the"wrong side," to use Vandenbroucke'sterms; he was unwilling to seriously examinethe data and to review all the evidence beforehim to try to reach a judicious conclusion.

In this article, I briefly review Fisher's lifeand times and then present his own dataand arguments. Readers can then judge forthemselves if these are "textbook classics fortheir impeccable logic" (Vandenbroucke'sphrase), or if they are much less.

Who was Fisher? What was his back-

416

R. A. Fisher and the Lung Cancer Controversy 417

ground? What were his contributions to sta-tistics? How did he get interested in smokingand cancer? What were his arguments? Whydid he refuse to perceive the causal natureof the association?

FISHER'S BACKGROUND

The problem that confronts the historianis that most of what we know of Fishercomes from the lengthy biography writtenby his adoring daughter, Joan Fisher Box(2). She also wrote the biography of Fisherthat appears in the Encyclopedia of Statisti-cal Sciences. Married to George Box, afamous statistician at the University ofWisconsin, she sides with Fisher in his manycontroversies with Karl and Egon Pearson,with Neyman, and with Gossett. Her ac-count of his involvement with the lung can-cer controversy is typically uncritical andone-sided. This historical resource is valua-ble but must be taken for what it is, a daugh-ter's loving account of her famous and dom-ineering father.

R. A. Fisher was born in a suburb ofLondon in 1890 and died in Adelaide,Australia, in 1962 of colon cancer. His fatherwas a fine arts auctioneer who worked in theWest End of London, but his business failedwhen Fisher was 16, so economic circum-stances were modest. In 1904, when he was14, his mother died of peritonitis; that sameyear he won a scholarship in mathematicsto Harrow, the famous private academy.There he demonstrated his talent early onby winning a math medal.

At age 19, Fisher was awarded a mathe-matics scholarship to Cambridge Universityand in 1912 at 22 was made a "Wrangler,"a high academic achievement. He studiedstatistics and physics under Jeans. In 1912he published his first paper wherein the con-cept of maximum likelihood was intro-duced. He corresponded with Gossett (whopublished under the pseudonym of "Stu-dent") about some of his pioneering work.He was very interested in evolutionary the-ory and genetics and combined this interestin the eugenics movement, now discredited,

but very avant-garde before World War I. In1911, Fisher helped form the CambridgeEugenics Society and gave a speech at thesecond annual meeting. He was concernedthat the birthrate fell by social class, withthose of the highest class having the fewestchildren. To the eugenicists, this impliedselection against "qualities" thought to bringsuccess and achievement in life. They be-lieved this would lead to "degradation of thegenetic stock" and, thus, the decline of civi-lization. The connection between eugenicsand racist theories has been pointed out bymany authors, most recently by Kevles (3).To encourage those with excellent "quali-ties" to produce offspring, Fisher urged thepassage of a family allowance. The argumentwent as follows. The most intelligent figureout that they can improve their opportuni-ties in life and their economic conditions byhaving fewer children; the less intelligentdon't understand this and have more chil-dren. Fisher proposed as a solution a statesubsidy of the intelligent to encourage themto procreate.

Fisher was active in the London EugenicsSociety for 20 years and had many childrenhimself.

FISHER'S CAREER

In 1913 he took a job as a statistician witha London bank and trained with the Terri-torial Army. At the outbreak of war in Au-gust 1914, he volunteered immediately tofight for Great Britain. Rejected for badeyesight, he served his country during thenext 5 years by teaching high school physicsand mathematics, a job he disliked but con-tinued, according to his daughter, as a publicservice. One wonders what would have hap-pened (or rather, not happened) to the fieldof statistics if he had entered the infantryand been killed, as were nearly a millionyoung British men.

In 1917 he married Ruth Guiness andimmediately started a family. His incomewas limited, and he farmed to provide forthe family.

About the time of World War I, the"dean" of all English statisticians was Karl

418 Stolley

Pearson. He took notice of Fisher's workand genius and, as editor of Biometrika,arranged to publish some of Fisher's articles.Pearson published Fisher's paper describingthe general sampling distribution of the cor-relation coefficient. When Pearson pub-lished another article by Fisher about maxi-mum likelihood and editorially criticized itwithout first informing Fisher he would dothis, Fisher developed a strong antipathy forPearson, the first of Fisher's several feuds.

In September 1917, Fisher started work asa statistician at Rothamsted experimentalagricultural station which, under his leader-ship, was to become a world center for thetheoretical development of experimental de-sign. There he developed the analysis ofvariance, the principle and contribution ofrandomization, and the idea and importanceof replication. He made great contributionsto the understanding of confounding andcreated designs to handle problems createdby confounding. In 1925 he published Sta-tistical Methods for Research Workers (4),and 10 years later The Design of Experi-ments was published (5). In 1938 he andFrank Yates brought out Statistical Tablesfor Biological, Agricultural, and Medical Re-search, still used today (6). (See figure 1.)

Following up his work on the distributionof the correlation coefficient, Fisher derivedthe sampling distributions of other statisticsin common use, including the F distributionand the multiple correlation coefficient. Hedeveloped the theory of estimation in 1922.In later years he made many other contri-butions to genetic and evolutionary theorythat are considered central to the under-standing of the theory of natural selection(7).

Fisher was offered the chair as the GaltonProfessor of Eugenics at University Collegein London. Actually, a new Department ofEugenics was created in order to attract himto the University. Fisher would never haveagreed to work in the statistics departmentunder Karl Pearson because of the antipathybetween them which had originated withPearson's critical editorial in Biometrika.Consequently, two departments doing thesame kind of work coexisted at the Univer-

RGURE 1. Passport photograph of Ronald AylmerFisher at age 34. Reprinted from Box JF. RA Fisherthe life of a scientist. New York: John Wiley & Sons,Inc., 1978.

sity College—Statistics under E. G. Pearson,who headed the department after his father,and statistics (misnamed Eugenics) underFisher. An intense rivalry and bad feelingexisted between Pearson and Fisher whichwas reflected in their departmental activities.

Jerzy Neyman joined Egon Pearson inStatistics in 1934 and immediately chal-lenged some of Fisher's ideas on hypothesistesting, introducing the ideas of power anddecision theory which he developed furtherin the United States with Abraham Wald.Fisher was unaccustomed to being contra-dicted and confronted Neyman as follows(related to Constance Reid by Neyman whenhe was an old man in working retirement atthe University of California, Berkeley):

And he said to me that he and I are atthe same building... he had published abook and that's Statistical Methods for Re-

R. A. Fisher and the Lung Cancer Controversy 419

search Workers—and he is upstairs fromme so he knows something about my lec-tures—that from time to time I mentionhis ideas... this is not acceptable to him.And then I said, "Do you mean that if Iam here, I should just lecture using yourbook?" And then he gave an affirmativeanswer. "Yes," that's what he expected.And I said, "Sorry, no, I cannot promisethat." And then he said, "Well, if so, thenfrom now on I shall oppose you in all mycapacities." And then he enumerated them:member of the Royal Society and so forth.There were quite a few. Then he left.Banged the door (8, p. 126).

Fisher continued to do ground-breakingwork in his new post and even became in-terested in a seroepidemiology of sorts whenthe blood group discoveries were an-nounced. In 1943 he assumed the chair inGenetics at Cambridge University. Duringthis period, he was estranged from wife andfamily and depressed by the death of his sonGeorge in World War II. Still, he continuedto publish important statistical papers, con-sult, and do original genetic research. Hebegan to receive many honors, was knightedby the Queen for his work, and finallyretired, at age 67, in 1957 (perhaps not co-incidentally, the same year he became em-broiled in the lung cancer/smoking contro-versy).

FISHER'S ARGUMENTS CONCERNINGLUNG CANCER

Fisher developed four lines of argumentin questioning the causal relation of lungcancer to smoking. I will first list these andthen briefly describe the evidence he pro-duced in support of these arguments.

1) If A is associated with B, then not onlyis it possible that A causes B, but it is alsopossible that B is the cause of A. In otherwords, smoking may cause lung cancer, butit is a logical possibility that lung cancercauses smoking.

2) There may be a genetic predispositionto smoke (and that genetic predisposition ispresumably also linked to lung cancer).

3) Smoking is unlikely to cause lung can-cer because secular trend and other ecologicdata do not support this relation.

4) Smoking does not cause lung cancerbecause inhalers are less likely to developlung cancer than are noninhalers (9).

Fisher sees the argument that lung cancercauses smoking as an essentially unsup-ported speculation. His view is best de-scribed in his own words:

The subject is complicated, and I men-tioned at an early stage that the logicaldistinction was between A and B, B causingA, or something else causing both. Is itpossible, then, that lung cancer—that is tosay, the pre-cancerous condition whichmust exist and is known to exist for yearsin those who are going to show overt lungcancer—is one of the causes of smokingcigarettes? I don't think it can be excluded.I don't think we know enough to say thatit is such a cause. But the pre-cancerouscondition is one involving a certain amountof slight chronic inflammation. The causesof smoking cigarettes may be studiedamong your friends, to some extent, and Ithink you will agree that a slight cause ofirritation—a slight disappointment, an un-expected delay, some sort of a mild rebuff,a frustration—are [sic] commonly accom-panied by pulling out a cigarette and gettinga little compensation for life's minor ills inthat way. And so, anyone suffering from achronic inflammation in part of the body(something that does not give rise to con-scious pain) is not unlikely to be associatedwith smoking more frequently, or smokingrather than not smoking. It is the kind ofcomfort that might be a real solace to any-one in the fifteen years of approaching lungcancer. And to take the poor chap's ciga-rettes away from him would be rather liketaking away his white stick from a blindman. It would make an already unhappyperson a little more unhappy than he needbe (10, pp. 21-2).

Fisher never produced any data or orga-nized any study to follow up on this implau-sible hypothesis. It is also noteworthy thatFisher was a smoker himself. Part of hisresistance to seeing the association may havebeen rooted in his own fondness for smokingand in his dislike of criticism of any part ofhis life.

Fisher's data concerning the genetics ofsmoking are sparse indeed. In two letters toNature, he presents some tables based oninformation he received from a Professor

420 Stolley

Verschuer of Germany and from Dr. EliotSlater of London. There are no further de-tails about these twin studies nor do weknow how the smoking histories are ob-tained, categorized, or analyzed. He presentstables that the reader is expected to take atface value with almost no information aboutthe study protocol or methods of investiga-tion. In a 1979 review article, Cook hassummarized the Fisher data in a single andbrief table (table 1)(11).

Fisher concludes from these data thatsmoking is genetically determined, at leastin part, but his discussion is surprisinglybrief.

.. .There can therefore be little doubtthat the genotype exercises a considerableinfluence on smoking, and on the particu-lar habit of smoking adopted... Suchgenotypically different groups would be ex-pected to differ in cancer incidence... (12,p. 108).

In a second letter to Nature he has moredata, again sparsely described but purportingto give the smoking habits of female twins

raised either together or apart (tables 2 and3). The 53 monozygotic twin pairs appearin both tables 2 and 3, but it is surprisingand unexplained that over one-half (51 per-cent) of the pairs were raised separately. Nodefinition of "raised separately" is given inhis letter to Nature {13).

As Vandenbroucke mentions in his arti-cle, Floderus et al. (14) using the Swedishtwin registry fail to support these data. Giventhe lack of detail Fisher provides, it is im-possible to account for the discrepancies.

The secular trend and ecologjc data willbe dealt with later in the article where Idiscuss the difficulty Fisher had in synthesiz-ing data and how he wandered out of hisarea of expertise.

His argument about inhalation is the onlyone he fully developed, but his use of thedata was devious and selective.

In their first case-control study, Doll andHill showed a slight difference in reportedinhalation by cases and controls, with con-trols reporting slightly more inhalation: 62percent among cases versus 67 percent

TABLE 1. Fisher's data on smoking habits of monozygotic and dizygotk: twin pairs*

MonozygoticDizygotic

Alike

33(65)f11(35)

Smoking habits

Somewhat aSke

6(12)4(13)

Different

12(23)16(52)

• Source: references 11-13.f Numbers in parentheses, percentage.

TABLE 2. Fisher's data on smoking habits of monozygotic and dizygotic female twin pairs*

MonozygoticDizygotic

•Source: references 11-13.t Numbers in parentheses, percentage.

TABLE 3. Fisher's data on smoking

Raised separatelyRaised together

Aike

44(83)t9(50)

habits of female

Affke

23(85)t21 (81)

Smoking habits

Not aike

9(17)9(50)

Total

53(100)18(100)

monozygotic twins raised together or separately*

Smoking habits

Not aike

4(15)5(19)

Total

27(100)26(100)

'Source: references 11-13.t Numbers in parentheses, percentage.

R. A. Fisher and the Lung Cancer Controversy 4 2 1

among controls. This was surprising, but intheir second case-control study this was notconfirmed, and when they combined infor-mation about inhalation in both studies, theresults were 67.5 percent inhaling amongcases versus 65.9 percent among controls—essentially no difference.

Fisher seized upon this discrepancy andmakes much of it. He uses a device thatsmacks of sophistry to magnify the differ-ence and its importance by transforming thepercentages into observed versus expectedfigures (using a chi-square analysis). He thensuggests that, if the cases had inhaled, 45lives could have been saved. He does thiswith attempts at humor and a mischievoustone that seemed to have pleased the audi-ences that attended his lectures on this sub-ject. The data are presented in the tablesbelow, in which I have summarized his data(tables 4 and 5).

Fisher interprets the data as follows:

.. .Should not these workers have let theworld know, not only that they have dis-covered the cause of lung cancer (ciga-rettes), but also that they had discoveredthe means of its prevention (inhaling ciga-rette smoke)? How had the Medical Re-search Council the heart to withhold this

information from the thousands whowould otherwise die of lung cancer?

Those who refuse the jump from theassociation to causation in the case of cig-arette smoking will not be tempted to takeit in the case of inhaling; but the MedicalResearch Council and its Statistical Re-search Unit think the argument is valid inthe first case. Can they refuse to admit it inthe second (10, p. 47)?

He used the data from only the first case-control study even though the data from thesecond study and the combined data wereavailable at the time; he probably did thisbecause the inhalation differences becameless after the data were combined. Using thistechnique, he has metamorphosized a trivialdifference between inhalers and noninhalersinto 45 lives saved if only the noninhalershad inhaled!

Other studies were available at the timethat showed no differences or showed moreinhaling by cases, but he never mentionsthis. Subsequent studies, of course, haveshown that the cancer risks are associatedwith inhalation.

Throughout his writings on the topic ofsmoking and cancer, he strives to be provoc-ative rather than responsible. Fisher fails todeal with the entire body of evidence on the

TABLE 4. Doll and Hill data from first case-control study of smoking and lung cancer, smoking prevalenceIn cases versus controls (men only)*

CasesControls

it

148179

1-14

Nt

84101

Maximum no.

15-24

I

133157

N

6344

of daily cigarettes

25-49

I

9674

N

7844

I

2116

>49

N

247

• Source: adapted from reference 11.11, inhaling history reported; N, nomnhaling reported

TABLE 5. Rsher's estimate* of lives saved if the cases not inhaling had been Inhaierst

Cigarettes/day

Expectedcancers

Observedcancers

No. of avoidablecancers

1-1415-2425-49>49

Total

14915310933

444

1491339621

399

0201312

45

• For detaBs of the computation, see reference 11.t Source: data adapted from reference 11.

422 Stolley

subject and makes too much of small differ-ences. While he claimed to be interested inthe dispassionate search for truth, his argu-ments have a polemic tone.

THE FAILURE OF HIS GENIUS

Why would Fisher persist in pressing thesearguments when the main body of data, evenby 1957-1958, was so large and firm in theindictment of cigarettes and his own coun-terarguments so tenuous and precious? Whydidn't he try to integrate or even weigh hisarguments against the rest of the data? Innone of his writings on this subject do wehave any attempt to evaluate the importanceof the inhalation exception and put it insome sort of proper relation next to the largebody of available epidemiologic data. Insum, he had trouble distinguishing the pos-sible from the probable and was unable tosynthesize all the data to reach a reasonableconclusion.

I believe, from trying to reconstruct hisinvolvement, by accounts of his contempor-aries of his personality, from his own writ-ings, from testimony of his daughter, and bysome guessing (which he would disdain),that there are several explanations for thefailure of his genius in this matter. Fisher'szest for confrontation and polemic was leg-end; all who knew him comment on this,even his usually uncritical biographer/daughter. Domestic fits of temper paralleledthe intolerance and aggressiveness he dem-onstrated toward his colleagues. He feudedwith the two Pearsons, Karl and Egon; withGossett; with Neyman; and with others. Hehated to admit that he was wrong on anysubject. He preached the scientific methodand was eloquent on the subject, but hadgreat difficulty following his own strictures.I have already quoted Neyman on Fisher'sattempt to force him to teach only Fisher'stheories and to censor him. There are alsoother accounts, even by those who reveredhim, as did Mather and Mahalanobis (7).Fisher described his own behavior relativelywell in 1947 when, reflecting on the way hefelt he was treated by Karl Pearson, he said:

A scientific career is peculiar in someways. Its raison d'etre is the increase ofnatural knowledge. Occasionally, therefore,an increase in natural knowledge occurs.But this is tactless and feelings are hurt. Forsome small degree it is inevitable that viewspreviously expounded are shown to beeither obsolete or false. Most people, Ithink, can recognize this and take it in goodpart if what they have been teaching for tenyears or so comes to need a little revision;but some undoubtedly take it hard, as ablow to their amour propre, or even as aninvasion of the territory they had come tothink of as exclusively their own, and theymust react with the same ferocity as we cansee in the robins and chaffinches thesespring days when they resent an intrusioninto their little territories. I do not thinkanything can be done about it. It is inherentin the nature of our profession; but a youngscientist may be warned and advised thatwhen he has a jewel to offer for the enrich-ment of mankind some certainly will wishto turn and rend him (2, p. 131).

This was written by the same man whotried to bully Neyman! When his first letterto the British Medical Journal was attackedand he was impugned for taking a fee fromthe tobacco industry, it probably fixed hisviews. His daughter mentions how offendedhe was by the rebuttal letters that pointedout he was a paid consultant.

Secondly, Fisher was a political conser-vative and an elitist (as were most eugeni-cists) and was disturbed by the British Med-ical Association's appeal to censure cigaretteadvertising and launch a public health cam-paign against smoking (Fisher was a smokerof pipes and cigarettes). He compares thisproposed public education campaign to to-talitarian propaganda and complains it ispremature in a letter to the British MedicalJournal:

Your annotation on "Dangers of Ciga-rette-smoking" [sic] leads up to the de-mand that these hazards "must be broughthome to the public by all the modern de-vices of publicity." That is just what someof us with research interests are afraid of.In recent wars, for example, we have seenhow unscrupulously the "modern devicesof publicity" are liable to be used under theimpulsion of fear, and surely the "yellowperil" of modern times is not the mild and

R. A. Fisher and the Lung Cancer Controversy 423

soothing weed but the organized creationof states of frantic alarm.

A common "device" is to point to a realcause for anxiety, such as the increasedincidence of lung cancer, and to ascribe itin urgent tones to what is possibly an en-tirely imaginary cause. Another, also illus-trated in your annotation, is to ignore theextent to which the claims in question havearoused rational skepticism. The phrase,"in the presence of the painstaking inves-tigations of statisticians that seem to haveclosed every loophole of escape for tobaccoas the villain in the piece," seems to be purepolitical rhetoric, even to the curious prac-tice of escaping through loopholes. I believeI have seen the source of all the evidencecited. I do see a good deal of other statisti-cians. Many would still feel, as I did aboutfive years ago, that a good prima facie casehad been made for further investigation.None think that the matter is already set-tled. The further investigation seems, how-ever, to have degenerated into the makingof more confident exclamations with thestudied avoidance of the discussion of thosealternative explanations of the facts whichstill await exclusion (15, p. 1518).

Fisher was upset by the public health re-sponse to the dangers of smoking not onlybecause he felt that the supporting data wereweak, but also due to his holding certainideologic objections to mass public healthcampaigns.

Third, it was recognized by those whoworked most closely with him that he wasgood with data while working on one smallset but was not easily able to integrate mul-tiple or large data sets. Here is a quotationfrom Yates and Mather:

In his own work Fisher was at his bestwhen confronted with small self-containedsets of data, and many of his solutions ofsuch problems showed great elegance andoriginality. He was never much interestedin the assembly and analysis of largeamounts of data from varied sources bear-ing on a given issue. The analysis of a singleexperiment and the conclusions that couldbe drawn from it, for example, interestedhim greatly, the assembly and analysis ofthe results of a varied collection of experi-ments scarcely at all. This would not havemattered—it could well be left to others—had he not tended to brush aside these morelaborious and pedestrian labours, while re-

membering and continuing to maintain hisown first conclusions based on an exami-nation of part of the data, conclusionswhich inevitably required reexamination inthe light of subsequent work.

The smoking-lung cancer controversy isa case in point. To those who mistrust thealarms and excursions of the medical worldFisher's skepticism of the evidence that cig-arette smoking had been established as thecausative agent of lung cancer was refresh-ing. His deduction from Doll and Hill'spublished results that inhalers were lesssubject to cancer than non-inhalers, andhis subsequent confirmation of this from amore detailed breakdown of the results sup-plied to him by Doll and Hill, was a strikingdemonstration of the pitfalls attendant onthe interpretation of survey data; and hisdemonstration, from data on monozygoticand dizygotic twins, that smoking habitsare strongly genetically conditioned was de-lightful. Yet although for some years pas-sionately interested in the controversy, henever attempted any review of the laterevidence (16, p. 106).

Perhaps another reason he persisted in thecigarette controversy may have had to dowith the circumstances of his life when hebecame engaged with the issue. He had justretired and was at loose ends. He received alot of public attention for his views on lungcancer and fashioned a talk on the subjectwhich he gave all over the world and whichhe arranged to have reprinted. It was wellreceived, even by medical audiences. Hisdaughter describes the lectures he gave onthis subject:

In Australia his lectures on the subjectwere frankly intended to raise questions.He spoke of "teasing" the British MedicalAssociation and of being "deliberately pro-vocative." He obviously enjoyed givingthese talks, which so clearly illustratedpoints he wanted to make about inference:the consequences of not randomizing, theneed to check on the assumptions implicitin any hypothesis, and the need to admitall of the data, particularly those whichseemed anomalous. Audiences were quickto laugh when he pointed out the negativecorrelation between lung cancer and inhal-ing; and there were other anomalies; thepositive correlation between smoking habitand genotype, the lack of success at thattime in isolating any carcinogenic sub-

424 Stolley

stance from tobacco, the difference in in-cidence of cancer between smokers of cig-arettes and smokers of tobacco in otherforms, between people in different environ-ments, between the sexes. It became ob-vious that more and better data wereneeded before any hypothesis could be con-sidered to have been convincingly demon-strated. If he seemed to some the devil'sadvocate, his reasonableness, good-humorand wit were hard to resist. His talks weregenerally well-received (2, p. 475).

Such a positive reception, as well as theopportunity to laud his personal achieve-ments and opinions, must have made thesetalks gratifying to a man nearing the end ofhis career.

Fisher's belief in the importance of ge-netics and his feeling that it is often neglectedin medical research also influenced hisviews; indeed, he says this in his discussionof the genetic predisposition to smoke. Hisadvancing the hypothesis as to the geneticpredisposition to smoke in an effort to sup-plant the cigarette hypothesis reflects his in-tense interest in genetics.

FISHER OVEREXTENDED

Perhaps the greatest contributor to hiserroneous analysis is that the great man hadwandered too far out of his field. He knewvery little of the case-control method butwas entirely suspicious of it because of theabsence of his beloved randomization. Norwas he well informed about the ecologic dataconcerning smoking and cancer and, whenhe comments on it, he gets it quite wrong.

The change over recent decades gives notthe least evidence of being due to increasingconsumption of tobacco. We can't tellmuch about the absolute magnitude of thissecular change. It is certain that radiologyhas facilitated the detection of lung cancerenormously, that radiological apparatusand radiologists are much more abundantlyavailable for our populations than they for-merly were.

. . .Again, the attention of the medicalprofession has been forcibly drawn to lungcancer, and it invariably happens that whenthe attention of the medical profession isdrawn to any disease, that disease begins totake up more space in the official reports—

it is more often seen and more often diag-nosed with confidence; death certificatesmore often include the particular disease.Consequently it is not easy to say howmuch of the increase is real. I think part ofit must be real, because there's no doubtthat the populations concerned have beenenduring or enjoying a very considerableincrease in urbanization.

. . .You might say that the whole popu-lation during the last twenty, thirty, fortyyears has been becoming steadily urban-ized, and as the urban rate for lung canceris considerably greater than the rural rate,in my country as in yours, we must recog-nize here the possibility of our real causeof the increase in lung cancer. There maybe others.

But the only good comparison we canmake in respect of the time-change is thatbetween men and women. The same ap-paratus, the same radiologists, the samephysicians diagnose both men and women.Whatever effects improved apparatus mayhave, whatever effects an increased atten-tion to the disease may have, will be thesame in the two sexes. Whatever effectsurbanization may have you would thinkmight be the same in the two sexes. Con-sequently, we can, at least, inquire whetherthe rate of increase of lung cancer in menis the same, or greater, or less, than the rateof increase of lung cancer in women. For itis certainly true, I think in both our coun-tries, that whereas the smoking habits ofmen have not changed very dramaticallyover the last fifty years, yet the smokinghabits of women have changed a very greatdeal. And on making that comparison, itappears that lung cancer is increasing con-siderably more rapidly—absolutely and rel-atively—in men than it is in women,whereas the habit of smoking has certainlyincreased much more extensively inwomen than in men. There is, in fact, noreasonable ground at all to associate thesecular increase in lung cancer with theincrease in smoking as has been done withdramatic eloquence, I suppose as part ofthe campaign of bringing home the terribledanger, just as though it was impossiblethat statistical methods of inquiry shouldsupply a means of checking that very rashassumption (10, p. 24-5).

He apparently made no attempt to put allthe evidence together, to read all the olderpapers, to wonder if the disease was uncom-mon among populations that didn't smoke.He wasn't an epidemiologist and seemed not

R. A. Fisher and the Lung Cancer Controversy 425

to know how to approach the question ofthe causal connection of smoking and lungcancer. This resulted in his ignoring or nevermentioning important supportive data forthe causal hypothesis and concentrating ex-cessively on the exceptions or discrepantdata that he enjoyed discussing in public. Inshort, he was so delighted at the finding ofcounterevidence that he exaggerated its im-portance. In addition, he took a fee from thetobacco industry, although those who knowhim best doubt that the fee mattered verymuch. It was probably not large, and he mayhave stopped taking fees when it was re-vealed in the British Medical Journal letterscolumns. But we have no hard evidence onany of this.

CONCLUSION

I suppose this story illustrates the melan-choly fact that even a genius can lack insightand fail to develop a sense of proportionabout evidence. There is a type of mind thatis inclined to commit the fallacy of the pos-sible proof; i.e., because an explanation ispossible, it becomes somehow probable intheir minds, particularly if they thought ofit (17).

It is not surprising that Fisher and othersquestioned the association between smokingand lung cancer, controversy is fundamentalin keeping scientific investigation from stag-nating. What is startling is that someone ofFisher's intellectual caliber would allow iso-lated evidence which coincided with his pre-viously held views to blind him to all else.Fisher's persistence and stubbornness ofopinion, even in the face of mounting evi-dence to the contrary, clearly demonstratehow even genius can err when it allows the

standards applied to the evaluation of evi-dence to weaken or vary according to theissue.

REFERENCES

1. Vandenbroucke JP. Those who were wrong. Am JEpidemiol 1989; 130:3-5.

2. Box JF. RA Fisher the life of a scientist. New York:John Wiley & Sons, Inc, 1978.

3. Kevles DJ. In the name of eugenics: genetics andthe uses of human heredity. New York: Alfred AKnopf, Inc, 1985.

4. Fisher RA. Statistical methods for research work-ers. Edinburgh, Scotland: Oliver and Boyd, 1925.

5. Fisher RA. The design of experiments. Edinburgh,Scotland: Oliver and Boyd, 1935.

6. Yates F, Fisher RA. Statistical tables. Edinburgh,Scotland: Oliver and Boyd, 1938.

7. Mahalanobis PC. Biography of RA Fisher. In: RAFisher. Contributions to mathematical statistics.New York: John Wiley & Sons, Inc, 1950. (Re-printed from Sankhya 1938;4(2):265-72).

8. Reid C. Neyman—from life. New York: Springer-Verlag New York, Inc, 1982.

9. Fisher RA. Smoking—the cancer controversy;some attempts to assess the evidence. (Pamphlet).Edinburgh, Scotland: Oliver and Boyd, 1959.

10. Fisher RA. Cigarettes, cancer and statistics. CentenRev 1958;2:151-66.

11. Cook RD. Smoking and lung cancer. In: FienbergSE, Hinldey DV, eds. RA Fisher, an appreciation.(Lecture notes in statistics: 1). New York: SpringerVerlag New York, Inc, 182-91.

12. Fisher RA. Lung cancer and cigarettes? (Letter).Nature 1958; 182:108.

13. Fisher RA. Cancer and smoking. (Letter). Nature1958; 182:596.

14. Floderus B, Cederlof R, Friberg L. Smoking andmortality: a 21 -year follow-up based on the Swedishtwin registry. Int J Epidemiol 1988; 17:332-40.

15. Fisher RA. Alleged dangers of cigarette smoking.(Letter). BMJ 1957;2:1518.

16. Yates F, Mather K. Ronald Aylmer Fisher, 1890-1962. In: Biographical memoirs of the fellows ofthe Royal Society of London. Vol 9. 1962:91-120.(As reprinted in Bennett JH, ed. Collected papersof R. A. Fisher. Vol 1. Adelaide, Australia: Univer-sity of Adelaide, 1971:106.)

17. Fischer DH. Historian's fallacies: toward a logic ofhistorical thought. New York: Harper & Row, Pub-lishers, Inc, 1970.