A Review of "Smoking and Health"

Author(s): K. A. Brownlee
Source: Journal of the American Statistical Association, Vol. 60, No. 311 (Sep., 1965), pp. 722739
Published by: American Statistical Association
Stable URL: http://www.jstor.org/stable/2283241 .
Accessed: 25/02/2011 17:57
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 service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Statistical Association is collaborating with JSTOR to digitize, preserve and extend access to Journal
of the American Statistical Association.

http://www.jstor.org

A REVIEW OF "SMOKING

AND HEALTH"*t

K. A. BROWNLEE
University
of Chicago
ALTHOUGHtherewereearlierpapers in the medical literature,e.g. Mtiller [I],

the possibilityof an associationbetweencigarettesmokingand lung

I
generallywith
cancerfirstbecame commonknowledgeamongstatisticians
in 1952.A retrospective
study
paperpublished
Doll and Hill's [2] retrospective
is one in whicha groupof,e.g.,cancerpatientsis classified
by theirsmoking
knownto nothavecancer,butsampled
habits,anda secondgroup,thecontrols,
fromthe same populationas the cancerpatients,is also classifiedby their
2 X 2 table,in whiclh
in principle
thetwosample
habits.The resulting
smoking
rrhe evidencefortheassociation
sizesarefixed,can be testedforindependence.
withthepublicationin 1954oftworeportsolnprosbecamemoreconvincing
pectivestudies,one in Englandby Doll anidHill [3] and oniein the United
study,a populationis
States by Hammondand Horn [4]. In a prospective
is
classified
his
smoking
habitsand
e.g.,
then
each
individual
by,
and
sampled
to
withrespectto cancerafteran intervaloftimesufficient
by his experience
allowa reasonablenumberofcasesofthediseaseto develop.
a wistfulness
thatstatisticshas
I have observedamongstsomestatisticians
Sincethisassociationbenotso farplayeda largerpartin sciencegenerally.
bymanyas oneofcausation,is of
and lungcancer,interpreted
tweensmoking
one mighthave expectedit to be greetedwithenthusiasm
primeimportance,
thecomments
and reactionsofthestaOn thecontrary,
and loud admirationi.
restrained.
By and large,in fact,thesilence
have beenveriy
tisticalprofession
has beendeafening.
studies,
In 1955,Neyman[5] pointedout a possiblefallacyin retrospective
had
no
direct
information
on the
in
a
that
footnote
he
asserted
but carefully
contains
and
cancer.
this
footnote
Incidentally,
between
smoking
association
the curiousremark:
inTable 1
ofthefigures
"A referee
warnsmethatinspiteofthefictitiousnless
of
the
character
on
remarks,
in
the
my
methodological
emphasis
and spiteof
it
and
for
the
use
publicity
argument
the 'tobacco people' may pick up
"
purposes.
Alsoin 1955,Berkson[6] anidMainlandaindHerrera[7] pointedoutcertain
studies.
prospective
typesofbiaswhichcouldaffect
His overallconclusionat that
the
literature.
reviewed
Cutler
In 1955,
[8]
the
evidenceat hand warrantsa
is
whether
"There
disagreement
timewas
are
cancer
and
that
causallyrelated."He also resmoking lunig
conclusion
in smokers
morefrequently
occur
cancer
should
"If
consistently
marked lung
the
thecase
in
of
in
various
studied,
populations
than non-smokers,
subgroups
* An invitedreviewarticleon "Smokingand hlealth,I Reportofthe AdvisoryCommitteeto the SurgeonGeneral ofthe Public Health Service,U. S. DepartmentofHealth,Education,and Welfare.Public Health ServicePubof Documents,GovernmentPrintingOffice,Washington,D. C. 20402. xvii,387.
licationNo. 1103. Superintendent
$1.25. Paper.
and W. H. Kruskalforcommentstheymade
t I wishto thankJ. Berkson,L. A. Goodman,G. W. Haggstromn,
on an earlierdraftofthisreview.I also wishto thankseveralrefereesfortheirsuggestions.

722

A REVIEW

OF "cSMOKING AND HEALTH"

723

To refuteeffectively
relationshipwillbe greatlystrengthened.
fora cause-effect
the hypothesisthat smokingis a cause of lung cancer would then require a
reasonable explanation, other than causation, for the consistentlyobserved
associationbetweensmokingand lung cancer."
In 1957, Sir Ronald Fisher [9] offeredsuch an explanation: "that cigarette
smokingand lung cancer,thoughnot mutuallycausative, are both influenced
by a commoncause, in this case the individuialgenotype."
Berksonin a series of papers, e.g. [10] and [11], expresseddoubt regarding
the demonstrationof causation. He emphasized that a higherdeath rate was
the excess
shownforalmostall diseases,and that ifthisis accepted as reflecting
deaths due to smokingthen about 40 per cent of all deaths among cigarette
smokersmust be attributedto theirsmoking.
In 1959 Cornfieldet al [12] concludedthat "the consistencyof all the epidemiologicand experimentalevidencealso supportsthe conclusionof a causal
relationship(of lung cancer) with cigarettesmoking,while there are serious
inconsistenciesin reconcilingthe evidence with otherhypotheseswhich have
been advanced."
Of the commentsof statisticians,the above seem to be among the more
important.
In 1962 the SurgeonGeneralofthe Public Health Serviceofthe U. S. Department of Health, Education, and Welfaremoved to appoint a committeeto
"assess available knowledgein this area [smokingvs. health] and make appropriate recommendations"(page 7). It is stated (page 8) that the functionof
this committeewas to make "an objective assessmentof the nature and magnitude of the health hazard" and "this committeewould produce and submit
to the Surgeon General a technical report containing evaluations and conclusions." However, "Recommendationsfor action were not to be a part" of
the committee'sresponsibility.
The formationof the committeewas a formidableoperation. The Surgeon
General met on July 24, 1962, with representativesof the American Cancer
Society,the AmericanCollege of Chest Physicians,the AmericanHeart Association, the AmericanMedical Association,The Tobacco Institute,Inc., the
Food and Drug Administration,the National Tuberculosis Association, the
Federal Trade Commission,and the President's Officeof Science and Techthat no statisticalsocietywas represented.Also the
nology. It is notewortlhy
list of organisations appears to be heavily weighted towards government
agenciesand organisationslarge,general,and active in public relations,and to
have low representationof societies with specificallyscientificoutlooks. The
participantsof the July24 meetingcompileda list of 150 scientistsand physicians. This list was then screened by these participants,each organisation
representedhaving a power of veto. From the intersectionof all the sets of
non-vetoes(which might,but apparentlydid not, have turnedout to be the
emptyset,thoughwe are not told its size) ten wereselectedforthe Committee.
Of the ten, only one is listed as a statistician.Since a substantialfractionof
the relevant issues are statistical,one might question whethera one in ten
representationfor statisticswas sufficient.However, one furthermemberof
the Committeeis listed as an epidemiologist.

724

AMERICAN

STATISTICAL

JOURNAL ASSOCIATION,

SEPTEMBER

1965

The committeemet for nine sessions of fromtwo to four days between

November 1962 and December 1963, and in addition therewere "uncounted"
meetingsofsubcommittees.The committeehad the assistanceof a professional
staffof eight,with a secretarialand technicalstaffof fifteen.The committee
lists 189 individualsor organisationswho made "contributions."The literature
surveyedwas enormous.A basic bibliographyof morethan 6000 articlesfrom
1200 journals up to 1959 was supplementedby an additional 1100 titles provided by the National Libraryof Medicine. The Report lists in the references
to its various chapters about 900 references.Some of the individual studies
consideredwere themselves enormous. For Hammond's latest report [13],
68,116 volunteersenrolled1,078,894men and women.
The Reportrepresentsa tremendoussifting,by a large group,ofan enormous
literatureinto 387 pages, and this reviewwill confineitselfto a briefsummary
of its contentsand an assessmentof its major conclusions,with emphasis on
statisticaland methodologicalaspects.
The frontmatter contains no trace of a statementof date of publication.
The very last page (page 387) does have as a footnotethe notation "U. S.
GovernmentPrintingOffice:1964 0-714-422" in which I presumethe "1964"
is the year of publication.
Chapter 1 reviewsthe evolutionof interestin the possible deleteriouseffects
ofsmokingand the proceduresforthe establishmentofthe Committee.Chapter
2 describesthe operationof the Committeethroughits subcommittees,consultants,etc. Chapter 3, "CriteriaforJudgment,"discusses"the Epidemiologic
Method." It remarks(pages 20-21) "Statisticalmethodscannotestablishproof
of a causal relationshipin an association. The causal significanceof an association is a matter of judgment. . . To judge or evaluate the causal sigof the association betweenthe attributeor agent and the disease, or
nificaince
effectupon health, a numberof criteriamust be utilized,no one of which is
an all-sufficient
basis forjudgment.These criteriainclude:
(a)
(b)
(c)
(d)
(e)

The
The
The
The
The

consistencyof the association

strengthof the association
specificityof the association
temporalrelationshipof the association
coherenceof the association"

The Report does not definethesetermsat thispoint,thoughtheyare discussed

in pages 182-189 in connectionwith lung cancer. The Report does not state
whetherthese are jointlysufficient.
Nor does this discussionof "the epidemioas discussed by
logic method" give any mentionof the factorof self-selection,
Yerushalmy [14], Yerushalmyand Palmer [15]. Later in the Report (pages
180-181), there is a mentionof selection bias, both by the operator of the
survey and by the individual whose cooperationis being sought. The subsequent discussion (page 181, last paragraph) is concernedwith the results of
selectionofthe type caused by exclusion,or decreasedprobabilityof inclusion,
of personssick or about to be sick: the discussiondoes not bear on the wellknown fact that volunteers(or equivalently,cooperative persons) may, and
oftendo, differin theirdisease experience.

A REVIEW

OF "SMOKING

AND HEALTH"

725

Chapter 3 also has a section on "Causality," (pages 20-21), the essential

part of which appears to be "It is recognizedthat oftenthe coexistenceof
several factorsis requiredforthe occurrenceof a disease, and that one of the
factorsmay play a determinantrole, i.e. withoutit the other factors. . . are
impotent. . .
The word cause is the one in generalusage . .. and is capable of conveying
the notion of a significant,effectual,relationshipbetween an agent and an
associated disorderor disease in the host."
Presumablyin the lung cancer situationcause must be definedprobabilistisketch might be as follows.Suppose that each
cally. A very over-simplified
individual and his circumstancesand environmentcan be classifieddichotomously on each of n classifications,so that he is eitheran E, or an Eic, an E2
or an E2c,etc., whereE1c is the complementof E1, etc. Suppose now that
Pr{E,I E2EJ * * * Ek} z Pr{E,I E2CEj . . *Ek},
whereE .*. . Ek is some specifiedsubset of conditionsE3, * * * , E.. Then, for
that set of conditionsEj . .. Ek, the probabilityof E1 given E2 is different
fromthe probabilityofE1 givenE2C.Since the presenceor absence ofE2 affects
the probabilityof E1, it seems reasonable to say that E2 is a cause of E1. The
practical significanceof E2 as a cause of E1 will depend on two conditions:
(a) If the lefthand side of the above inequality,thoughnot exactly equal
to the righthand side, is numericallyquite close,thenforthat set ofEj Ek
the causative associationmay be of small practicalimportance.
(b) If the inequalityholds only for,say, one particularsubset Ey, * Ek,
and forall othersubsets equality holds, and if the subset Ej, * * * , Ek occurs
in the populationwithlow probability,
then Pr{El IE2 }, whilenot strictly
equal to Pr I E, IE2c},willbe numericallyclose to it, and then E2 as a cause of
E1 may be of small practical importance.These considerationsare related to
the Committee'sresponsibilityfor assessmentof the magnitudeof the health
hazard (page 8). Furthercomplexitiesarise whenwe distinguishbetweencases
in which one of the requiredsecondaryconditionsEj, * * *, Ek is, on the one
hand, presumablycontrollableby the individual,e.g. the eating of parsnips,
or uncontrollable,e.g. the presenceofsome geneticproperty.In the lattercase,
it furthermakes a differencewhetherthe genetic propertyis identifiableor
for example, it could be brown eyes which is the significant
non-identifiable:
subsidiaryconditionE1, and we could tell everybodywith not-browneyes it
was safe forthemto smoke.
Chapter 4 is a set of Summariesand Conclusions.Chapter 5 gives consumption of tobacco products,includingchewingtobacco and snuff,forthe United
States for selected years from1900 to 1962. Chapter 6 is on "Chemical and
Physical Characteristicsof Tobacco and Tobacco Smoke." Chapter 7, "Pharmacologyand Toxicology of Nicotine," concludes that nicotineis unlikelyto
be important.
In Chapter 8 we come to a discussionof the relationshipbetweensmoking
and mortality.Subsequent chapters deal with specific groups of diseases:
9, Cancer; 10, Non-Neoplastic Respiratory Diseases; 11, Cardiovascular
Diseases; 12, OtherConditions.Chapter 13, "Characterizationof the Tobacco

726

AMERICAN

STATISTICAL

ASSOCIATION

JOURNAL,

SEPTEMBER

1965

Habit," concludesthat smokingis a habituationratherthan an addiction,"since

once establishedthereis littletendencyto increasethe dose," etc. (page 354).
Chapter 14, "Psycho-Social Aspects of Smoking," states that "The overwhelmingevidence points to the conclusionthat smokingis to a large extent
psychologicallyand socially determined"(page 377). The "psychologically"
part of this conclusionis apparentlybased on Eysenck [16] who found that
cigarettesmokiingwas positivelyassociated with extroversion,and that both
non-smokersand pipe smokers(the studywas made in England, so presumably
therewere too few cigar smokersto consider)were less extroverted.
The evidenceforthe "socially"part ofthe conclusionseemsin its summarized
formto be not veryquantitative.The Reportformsthe conclusionthat "white
collarprofessional,managerialand technicaloccupationscontainfewersmokers
than craftsmen,salespersons,and laborers" (page 363), but does not give any
figures.However, "As to separate class-linkedvariables,incomedoes not seem
to be related in a consistentmannerto prevalence of smoking."The Report
finds"The relationshipbetweensmokingand education is unclear," and "The
proportionof smokersis roughlythe same among whites and non-whites."
Men smoke morethan women,thoughthe discrepancyhas been decreasingin
recent years. Also, "Smoking (of any kind) is most prevalent among the divorced and widowed and least among those who have never been married,
except that among personsover 45, never marriedare as likelyto be smokers
as the married" (page 364). Some religionsaffectsmoking,and the ruralfarm
population smoke less than the rural non-farmpopulation. There is little differencebetween the latter and the urban population. Few figuresare quoted
in this discussion,so one cannotassess how importantthesevariousfactorsare.
Chapter 15 deals with the morphologicalconstitutionof smokers. What
is fromSeltzer [17], who in 1942 obtained
seems the most usefulinformation
anthropometricmeasurementson 922 Harvard undergraduatesand in 1959
ascertainedtheirsmokinghabits (withan 81 per cent response).He foundthat
cigaretteonly smokerswere statisticallylargerin a numberof anthropometric
indices than non-smokers,averaging4.37 pounds heavier, pipe only smokers
larger still, and cigar smokersthe largest. The Report concludes,however,
"The available evidence suggeststhe existenceof some morphologicaldifferences between smokersand non-smokers,but is too meagre to permita conclusion" (page 387). The Report does not give reasons fordecliningto accept
Seltzer's results.
We turn now to a detailed considerationof Chapters 8 and 9.
Chapter 8 reviewsthe seven major prospectivestudies and concludes that
the overall death rate for cigarettesmokersis about 70 per cent higherthan
fornon-smokers.For cigar and pipe smokersthe effectwas minor.The mnortality ratio, i.e. the ratio of the death rate forcigarettesmokersto the death
rate fornon-smokers,was particularlyhigh forcertaindiseases, e.g. cancer of
the lung (10.8), bronchitisand emphysema (6.1), etc. For coronaryartery
all causes of
disease the mortalityratio was 1.7. It seems that foreffectively
death (even accidents, suicide, and violence) the cigarettesmokershad the
higherdeath rate. Though cancer of the lung had the highestmortalityratio,
arterydisease is the chiefcontributorto the excessnumberofdeaths
"(coronary

A REVIEW

OF "SMOKING

AND HEALTH"

727

of cigarettesmokersover non-smokers"(page 113). This chapter makes no

explicitclaim that cigarettesmokingis the cause of the mortalityratios being
greaterthan 1.
Chapter 8 reaches its conclusionssolely on the basis of seven prospective
studies,presumably(thoughnothingis stated) forthe reason that its authors
wereaware ofthe hazards in attemptingto make rigorousinferences
fromretrospective studies (possible biases in retrospectivestudies are discussed in
it seemsto me,is that it is very
Chapter9, pages 180-181). The main difficulty,
to defineaccuratelythe populationfromwhichthe hospitalsample of,
difficult
e.g., cancerpatientshas been drawn,and even moredifficult
to draw a sample
of controlsfromthat population.
In some of the prospectivestudies a clearlydefinedpopulation existed,e.g.,
policyholdersof U. S. GovernmentLife Insurance Policies, and the intention
was to take a 100 per cent sample of this population,but in generalforthese
studiesnon-responseaveraged 32 per cent. Biases in the non-respondents
could
play havoc withinferencesbased on the respondents.AppendixI to Chapter 8
considersthis, and on various assumptions about the behavior of the nonrespondentsconcludes that such biases could, under ratherextremeassumptions,make an actual mortalityratio of, forexample, 4.0 be observed as 5.0,
or an actual 7.0 be observedas 10.0. The possibleeffectsofthe non-respondents
are tricky,however,as Doll found (page 97) that forBritishdoctorsthe nonrespondentshad a higherdeath rate and relativelymore smokersthan the
respondents.The cross-classification
is, however,not given.
In other prospectivestudies, e.g. the 1963 Hammond study, there is no
clearlydefinedpopulationand hence the conceptof percentageresponseis impossible to evaluate. Bias could exist in the recruitmentintothe sample, but
it is stated (page 181) that its effects,if any, should decrease with time, and
thisis not observed.Mainland and Herrera [7] had previouslywonderedabout
the effectsof biases in studiesof this type,but the Report cites the gradual increase in the mortalityratio with time as weakeningtheircriticism.The 1963
Hammondstudyspecificallyexcluded"personstoo illto answera questionnaire"
(also illiterates,and persons who could not have been traced). Presumably
theseexclusionsare responsibleforthe age adjusted death rate ofthe Hammond
sample being substantiallybelow that of U. S. males generally.In fact,forall
seven prospectivestudiesthe samples are healthierthan U. S. males (page 95).
This may be largelydue to the probable fact that the surveysare drawnfrom
relativelyhigh socio-economicgroups. In specificsurveysother reasons may
also operate.
Even thoughcancer contributesonly 26 per cent to the total excess number
of deaths of cigarettesmokers,it receivesthe most attentionin the Report, as
Chapter 9 being 136 pages is substantiallythe largest. Lung cancer, which
contributes16 per cent to the total number of excess deaths, receives the
greatestemphasis and we will confineour attentionaccordingly.
Chapter 9 firstreviews age-adjusted mortalityrates for various countries,
forvarious times (1900-1960), by site, by sex, income class, occupation, and
ethnicgroup. It then discussesvarious carcinogens,some of whichare present
in tobacco smoke. Then followspages 149-196 onlluilg cancer, reachingthe

728

AMERICAN

STATISTICAL

ASSOCIATION

JOURNAL,

SEPTEMBER

1965

conclusion(page 196) "Cigarettesmokingis causally relatedto lung cancer in

men." We will now summarizethe argumentsthat led to that conclusion.
the 7 prospective
Pages 150-161 review29 retrospectivestudiesand, briefly,
studiesthat had been discussedin Chapter 8. There followsa discussionof attemptsto induce lung cancer with tobacco extractsand tobacco smoke in experimentalanimals. These have not been successful.However, industrialcarcinogens were effective.Another section (page 167) reports that "genetic
factorsexerta determininginfluenceon the spontaneousdevelopmentand inductionoflung tumorsin mice."
Pages 175-176 discussthe correlationofnational crude male death rates for
lung cancer with per capita consumptionof cigarettes.Using a 20 year lag,
Doll obtained r=0.73 with 11 countries.The "population" fromwhich these
11 countrieswere "sampled" is not described.
Pages 179-182 discuss possible biases being responsiblefor the observed
association,and concludesthat the observedassociationis real.
The Report thenmovesto establishthe causal significanceofthe association,
its argumentsfollowingthe five criteriathat Chapter 3 listed as being part
(or all?) of the "epidemiologicmethod."
(a) The firstcriterionis "consistency." The 29 retrospectiveand the 7
prospectivestudies all demonstratethe same association.
(b) "The most direct measure of the strengthof the association between
smokingand lung cancer is the ratio of lung cancer rates forsmokersto non-,
smokers. . . " It is high,ofthe orderof9 to 1. Also "Importantto the strength
as well as to the coherenceof the association is the dose-effectphenomenon.
was
the dose-effect
In everyprospectivestudythat providedthis information,
apparent... "
(c) The specificityof the association "impliesthe precisionwith which one
componentof an associated pair can be utilized to predictthe occurrenceof
the other." This seems a curioususe of the word "specificity."A more satisfactorydefinitionwould seem to be Yerushalmyand Palmer's [15] "The basic
is that ifthe characteristicis not related
assumptionofsuch a testforspecificity
to the disease in a causal way, then the relationshipshould not be restricted
to the disease understudybut shouldalso be presentwithotherdisease entities.
If the characteristiccan be shown to be related only or mostlyto the disease
under study and not to many otherdisease entities,then our confidencethat
it is a cause-carryingvectorforthat disease is greatlyincreased."
its own and
This sectionof the Report discusses both these interpretations,
Yerushalmyand Palmer's, of "specificity."On the first,it commentsthat with
diseases with multiplecauses one cannot expect high specificityin this sense.
On the second, the Report states "The number of diseases in which the
high, afterconsiderationof the non-responsebias,
ratios remain significantly
is not great enough to cast serious doubt on the causal hypothesis."
It furthercommentsthat even a singlesubstancemightcause severaldiseases
and a mixtureof substances,such as occur in tobacco smoke,might"produce
more than a single disease." The Report concludes "Thus, it is reasonable to
conclude that the association between cigarettesmokingand lung cancer has
a high degree of specificity."

A REVIEW

OF "SMOKING

AND HEALTH"7

729

(d) The next section, "Temporal Relationship of Associated Variables,"

makes the point that people smoke and then have cancer, ratherthan vice
versa. The point of course,is that if B occurs later in time than A then it is
difficult
to maintainthat B is a cause ofA. This is obviouslya necessarycondition, but not sufficient.
(e) The finalcriterionis the "Coherenceof the Association"whichcovers a
numberof pointsas follows.
(1) The historicalincrease in cigaretteconsumptionis correlatedwith the
historicalincreasein lung cancer.
in smokingis in the same directionas the difference
(2) The sex difference
in lung cancer rates.
(3) Page 186 reportsthat "although adjustmentfor smokinghistorydoes
niotequalize the urban-rurallung cancer mortalityratio, controlon the
urban-ruralresidencefactorneverthelessleaves a large mortalityrisk
difference
betweensmokersand non-smokers."This seems to implythat
correlationof residencewith smokingcannot account forthe observed
associationoflung cancerrate withsmoking,and, vice versa, correlation
of smokingwith residencecannot account forthe observedassociation
of lung cancer with residence.
in lung cancer mortalityre(4) A section on socio-economicdifferentials
marks (page 187) "it willbe neverthelessnotedthat the professionaland
farmerand farmmanagergroupshad higherproportionsof non-smokers
among them than did the laborersand craftsmen.This findingis in the
proper directionforcompatabilitywith the socio-economicdifferential
in lung cancer mortalitybut the disparitydoes not appear to be sufficientto providea satisfyingcorrection."
(5) This section commentson the dose-responserelationshipand refersto
the muddled positionwith respectto inhalation.
A sectionon the histopathologicevidence(page 189) refersmainlyto workby
Auerbachon epithelialchangesin the trachea and bronchi:
"These changes were rarelyseen among non-smokers,but increasedin frequency and intensitywith the number of cigarettessmoked daily by individuals withoutcancer and were most frequentand intensein patients dying
of lung cancer." The opinionis expressed(page 172) "it seems probablythat
some of the lesionsfoundin the tracheobronchialtree in cigarettesmokersare
capable of developinginto lung cancer."
It is difficult
forone not a specialistin this area to assess the significanceof
these observations.The Report concludes (page 189) "Thus, the histopathologic evidence derived from laboratory and clinical material support the
cigarette smoking-lung cancer hypothesis" so the Committee does not
appear to regard this time as crucial. Presumablyhaving established to its
satisfactionthat smoking causes lung cancer since the criteriaof the "epidemiologicmethod" are allegedly satisfied,the Report in pages 190-193 discusses the "ConstitutionalHypothesis,"i.e. "the alternativehypothesisthat
both smokingof cigarettesand cancerof the lung have a commoncause . . . "
and concludes "that geneticfactorsplay a minorrole" (page 192). The main

730

AMERICAN

STATISTICAL

ASSOCIATION

JOURNAL,

SEPTEMBER

1965

reasons forthis conclusionappear to be (a) the presumptionthat the genetic

hypothesiswould have to be very complicatedto include the dose-response
relationship,etc., and (b) the rapid historicalrise of lung cancer.
Finally,sundryfactorssuch as occupationalhazards,urbanization,air pollution, etc., are mentioned,and on page 196 the conclusionof causality is unequivocallystated.
This reviewhas so farbeen largelyrestrictedto a summarizationof the relevant argumentsleading to the above importantconclusion.I will now discuss
the question of whetherthe available evidencejustifiesthe Report in reaching
that conclusion.It is possible of course,forthe conclusionto be in fact correct
but forthe evidencein its favorto be inadequate. We are all familiarwiththe
evidence,
accused who has to be found not guilty, because of insufficient
thoughwe are personallycertainof his guilt. Let us now commenton various
partsof the Report's position.
In my opinioii a key factorin determiningone's opinionsis one's interpretation of Table 26, pages 109-110, whichshowsthe mortalityratios,individually by separate study and also jointly,for25 causes of death, orderedfrom
cancer of lung with a total mortalityratio of 10.8, bronchitisand emphysema
witha total mortalityratioof6.1, downto cancerofthe rectum(1.0) and cancer
ofthe intestines(0.9). The lattercause of death is the onlyone out ofthe 25 to
show a total mortalityratio of less than 1. The total over all studies column
of Table 26 is reproducedhereas Table 1. The medianofthe 25 total mortality
ratios is 1.5, and 8 are greaterthan 2.0. One could adopt an extremenull hypothesisattitudeand admit that onlythose causes of death forwhichthe total
greaterthan 1 are genuinelyassomortalityratio was statisticallysignificantly
ciated with cigarettesmoking.The technicaldetails of this statisticalproblem
would be difficulton account of the various "sampling" proceduresinvolved,
allowances forpossible biases and dependencies,the varyingsample sizes, and
cause of
the multiplicityof the tests being made. If we take the twenty-first
death, "accidents,suicides,and violence," with a total mortalityratio of 1.2,
it is noteworthythat in each of the seven separate studies the mortalityratio
is greaterthan 1. In myopinionvirtuallyeverytabulated cause of death shows
cause of
a mortalityratio greaterthan 1. I do not regardeven the twenty-fifth
an exceptionto this, as if we
death, cancer of the intestines,as significantly
look at estimatesof 25 parametersthe smallestof the 25 is likelyto be smaller
than its population value, and at 0.9 it would not take much of a sampling
errorto bringthe ratio over 1. However, whetherone feelsthat probably all
25 causes of death have mortalityratios greaterthan 1, or merelythe first21
in my opinion. In the disout of the 25, or the first15, makes littledifference
cussion I give below, I shall use the phrase "virtuallyall causes of death" but
it could be replaced by some such phrase as "21 out of 25 causes of death"
withoutchangingappreciablythe importof the discussion.
One's interpretationof-Table 26 determineswhetherone accepts the Report's conclusion(page 185) "Thus, it is reasonableto concludethat the association between cigarette smoking and lung cancer has a high degree of
specificity"as, on the one hand, reasonable, or on the other hand, as an extraordinarystatement completelyin violation with the facts. The Report

A REVIEW

OF "SMOKING

731

AND HEALTH"

TABLE 1. NUMBERS OF EXPECTED AND OBSERVED DEATH FOR

SMOKERS OF CIGARETTES ONLY, AND MORTALITY RATIOS,
TOTALLED OVER ALL SEVEN PROSPECTIVE STUDIES. TAKEN
FROM TABLE 26, PAGE 110, OF THE REPORT
Cause of death
Cancer of lung
Bronchitis,emphysema
Cancer of larynx
Cancer of oral cavity
Cancer of esophagus
Stomachand duodenal ulcers
Othercirculatorydiseases
Cirrhosisof liver
Cancer of bladder
Coronaryarterydisease
Otherheartdiseases
Hypertensiveheartdisease
Generalarteriosclerosis
Cancer of kidney
All othercancer
Cancer of stomach
Influenza,pneumonia
All othercauses
Cerebralvascular lesions
Cancer of prostate
Accidents,suicides,violence
Nephritis
Rheumaticheart disease
Cancer of rectum
Cancer of intestines
All causes

Deaths
Expected

Observed

Mortality
rtio
ratio

170.3
89.5
14.0
37.0
33.7
105.1
254.0
169.2
111.6
6,430.7
526.0
409.2
210.7
79.0
1,061.4
285.2
303.2
1,508.7
1,461.8
253.0
1,063.2
156.4
290.6
207.8
422.6
15,653.9

1,833
546
75
152
113
294
649
379
216
11,177
868
631
310
120
1,524
413
415
1,946
1,844
318
1,310
173
309
213
395
26,223

10.8
6.1
5.4
4.1
3.4
2.8
2.6
2.2
1.9
1.7
1.7
1.5
1.5
1.5
1.4
1.4
1.4
1.3
1.3
1.3
1.2
1.1
1.1
1.0
.9
1.68

accepts "specificity"as one of the criteria of "the epidemiologicmethod"

(pages 20, 183), and in myopinionthe way it claimsthe factsare in conformity
withthe criterionis to flatlyignorethe facts.
whichthe Reporttotally
Table 26 raises difficulties
In my opinion,therefore,
are as follows.
failsto face up to. The difficulties
If one believes that the observed association between smoking and lung
cancer is substantiallyreal, and not an artefactof biased sampling,then one
would seem requiredalso to accept the observedassociation betweensmoking
and almost all causes of death as substantiallyreal.
If one believes that the observedassociation betweencigarettesmokingand
virtuallyall causes of death is substantiallyreal, thenone musttake one ofthe
followingpositions:
(a) All the associations are due to causation.
(b) Some of the associationsare due to causation and othersto correlations,
concealed or otherwise.
(c) All of the associationsare due to correlations,concealed or otherwise.

732

AMERICAN

STATISTICAL

ASSOCIATION

JOURNAL,

SEPTEMBER

1965

If one adopts position

Positions (a) and (c) may presentthe fewestdifficulties.
(a), then one is under some obligationto provide hypothesesas to possible
mechanisms,or to hold out hope that futureresearchwill provide these hywith the smokinghypothesisis that it has not really
potheses. The difficulty
got to firstbase on even lung cancer,let alone the other 24 causes of death,
even though the matterhas been under intensiveinvestigationforten years
or more.
The most potent carcinogenidentifiedin tobacco smoke is benz (a) pyrene
(page 27) and "it is presentin much larger quantity than any of the other
carcinogenslisted." Cigar smokehas almost 4 timesas much benz (a) pyreme,
and pipe smokeabout 10 timesas much as cigarettesmoke (page 58), yetpipes
and cigars are pretty well innocent of the charge of association with lung
cancer. (However, the Committeedoes conclude that the smokingof pipes is
causallyrelatedto cancerofthelip (page 204).)
Apart fromthe above specificmeasurementson smoke, the differencebetween cigarettesand cigars (and pipes) is a puzzlingfeatureof the indictment
of tobacco. It could be, of course,that the varietyof tobacco used differssigthat cigars and pipes do
nificantly,that the tobaccos are cured differently,
not use cigarettepaper, that possiblypipes and cigarsburn at lowertemperatures,or that many cigarettesmokersinhale whereasfewpipe or cigarsmokers
do. But if inhalationis the crucialitem,then it should show up stronglywhen
cigarettesmokersare analyzed into inhalersand non-inhalers,and as reported
above, the presentevidenceon this point is not clear.
As stated earlier,animal experimentsto demonstratecarcinogeniceffectsof
tobacco smoke have provedunsuccessful.
Whilehypothesesfora causative effectforlung cancerare stillunsatisfactory
aftera decade ofresearch,the situationis even emptierforotherdiseases.
The degreeof one's beliefin the plausibilityof the smokinghypothesisthus
is considerablyaffectedby the extent to which one thinksit probably that
satisfactorymechanismswill be provided to account for all, or most, of the
diseases.
If one adopts position(b), thenone is in the unenviablepositionofadmitting
that concealed correlationsaccount fordiseases Di, i = k, k+ 1, * * * , n, but are
not responsiblefordiseases Di, i== 1, 2, * . , k- 1. It seems to me that once
one admits that concealed correlationsaccount for a substantial number of
the observed associations, then one has to work very hard to disprove the
hypothesisthat they may account forall the observed correlations.
The Report uses the historicalcorrelationof rise in lung cancer mortality
with increasein per capita consumptionof cigarettesas one factorto support
the coherenceof the association (page 185). All statisticiansknow that the
presence of a positive, zero, or negative correlationbetween two variables
observed over time has been the basis of more ludicrous nonsense than any
otherstatisticalprocedure.For example,the incidenceofcancerofthe stomach
has been decliningformanyyears (Dorn and Cutler [18]), but onlya madman
would inferthat the increased smokinghas caused the decreased stomach
cancer. Incidentally,it is surprisingthat the Report makes no mentionof this
fact in the section on stomach cancer (pages 225-229) thoughit is mentioned
.

A REVIEW

OF "SMOKING

AND HEALTH"

733

on page 135. The Report omits any commenton the weakness of this item in
its discussionof the "Coherenceof the Association"on page 185.
Similar considerationsapply to spatial correlations,whichare presentedon
page 176.
The fact that the hypothesisthat cigarettesmokingis a cause of cancer
with the data, and hence that this
appears by and large to be in conformity
hypothesisis acceptable, does not rule out the possibilitythat thereare other
withthe data. As Sir Ronald Fisher [19] wrote
hypothesesalso in conformity
almost thirtyyears ago:
"For the logical fallacy of believingthat a hypothesishas been proved to
be true,merelybecause it is not contradictedby the available facts,has no more
rightto insinuateitselfin statisticalthan in otherkindsofscientificreasoning."
And as Yule [20], quoted by Irwin [21], wroteabout fortyyears ago: " 'You
can prove anythingby statistics'is a commongibe. Its contraryis morenearly
true-you can never prove anythingby statistics.The statisticianis dealing
with the most complex cases of multiple causation. He may show that the
facts are in accordance with this hypothesisor that. But it is quite another
thing to show that all other possible hypothesesare excluded, and that the
than the particularone he miay
factsdo not admit of any otherinterpretation
have in mind."
proceduresfor
It is not clear to me what are the optimal,or even satisfactory,
inference.In the case wherethere are two competinghypotheses,one might
estimatethe "plausibilityratio," analogous to the likelihoodratio, and if it is
verysmAll,<<1, or verylarge,>>1, one can reacha conclusion.If theplausibility
ratio is in the neighborhoodof 1, then no decision can be reached. One would
wishthat ifthe moreplausiblehypothesisis chosenas a null hypothesis,
further
thena test of thisnull hypothesiscan be accepted at a largeP value, forotherwise one would suspect that neitherhypothesiswas correct,and if the less
plausible hypothesisis tested as a null hypothesisone would want to it be rejected at a small P value.
The main alternativeto the smokingcauses-cancerhypothesisis the genetic
hypothesis,and there are several odd pieces of informationthat give plausibilityto it.
(1) There is some evidence that non-smokers,cigarettesmokers,and pipe
and cigar smokers are morphologicallydifferent(Chapter 15, referenceto
Seltzer [17]).
(2) There is some evidence that the various classes of smokers and nonsmokersare psychologicallydifferent
(Chapter 14, referenceto Eysenck [16]).
(3) There is some evidence that identical twins are more alike than nonidentical twins in their smokinghabits (page 190, referencesto Fisher [22],
Friberget al [13], and Raaschou-Nielsen [24]).
(4) "Foreign-bornmigrantsto the United States as a group have ageadjusted death ratesforcancerofthe esophagusand stomachabout twicethose
recordedfornative-bornwhite males and females,"etc., (page 134).
(5) "The several ethnicgroupsin the United States display theirown characteristicpatternsofsuccessesand deficitsin riskby site" ofcancer (page 135).
It is true,of course,that these phenomena((4) and (5)) could be the result

734

AMERICAN

STATISTICAL

ASSOCIATION

JOURNAL,

SEPTEMBER

1965

of social, dietary,or other customs importedby the immigrantsand maintained forone or moregenerations,or could be a confoundingof the effectsof
the environmentofthe regionofthe U. S. in whicheach ethnicgrouptendedto
settle. Thus these phenomena can be interpretedin terms of environmental
effectsratherthan geneticeffects.
(6) Yerushalmy [14] presenteddata stronglysuggestingthat whetheror not
the husband smokes is associated with the incidenceof prematurebirths.
(7) Hammond [13] gives resultsshowingmarkedassociation of death rate
withlongevityof parentsand grandparents.
On my reading,the main reasons the Report gives forrejectingthe genetic
hypothesisare (1) the necessary complexityof the genetic hypothesis,and
in the
(2) the unlikelihoodthat the geneticpool can have changed sufficiently
past fiftyyears to account forthe historicalrise in lung cancer rate.
genotypesneed to be hypothesizedforthe
different
For the firstdifficulty,
various classes such as cigarette smokers, cigarette smokers who give up
cigarettesmoking,cigarsmokers,etc. It seemsto me that an enormousnumber
oftraitsare transmittedgenetically,fromcolorofskinand tendencyto diabetes
and tendencyto baldness,and it is completelyplausible that the tendencyto
be a cigarettesmokerin varyingdegrees,a pipe or cigar smoker,and so on
can be carriedby the geneticcode.
The second objection referredto above seems to me to overlooka possible
ofhistoricalchangein death rates. Firstly,possibly
fallacyin the interpretation
part of the long term apparent rise may be an illusion caused by fashionsin
diagnosisand by improvementsin diagnostictechniques.It is quite probable
that several decades ago many cases of lung cancerwould have been diagnosed
as tuberculosis.The Report claims (page 140) that over the shorterterm,from
1947 on, particularlyfordata for Connecticutand New York, the increasein
genuine as there have been "no significantadvances in diagnosticmethods"
and in these regions"a highpercentageof the cases reportedhave microscopic
confirmation."This commentdoes not quite bear on the point at issue. The
suggestionof the skeptics is not that cases of tuberculosis are now being
falselydiagnosed as lung cancer but on the contrarythat in the past cases of
lung cancer werefalselydiagnosedas tuberculosis.
The second argumentis morecomplex.The generationbornin 1880 reached
the age of 20 in 1900, and a substantialfraction,24 per cent,had died by this
time,presumablylargelydue to the traditionalinfectiousdiseases of childhood.
The generationbornin 1900 reachedthe age of 20 in 1920,and a lesserfraction
ofthisgenerationhad died by that age, namely15 per cent.Therefore,the 1900
stratumfromthe 1880 cohort
cohortat the age of 50 representsquite a different
at the same age of 50, -sincethe formerincludes the "weaklings" who were
eliminatedfromthe latter.The relativelylarge numberof cases of lung cancer
observed in the 1900 cohortcan merelybe largelythose who would not have
survivedto run the riskof lung cancer if theyhad been born 20 years earlier.
On this model the alleged historicalincrease in rate of lung cancer can be
readily accounted for. The model is speculative,however.
Rather strangely,the Report does not include in the section devoted to refutingthe constitutionalhypothesis(pages 190-193) the observationsby Corn-

A REVIEW

OF "SMOKING

735

AND HEALTH"

TABLE 2

C
Cc
Sums

Sums over X and Xc

Xc

X
S

Sc

SC

SC

0.00450
0.44550

0.00050
0.04950

0.00005
0.04995

0.00045
0.44955

0.00455
0.49545

0.00095
0.49905

0.45000

0.05000

0.05000

0.45000

fieldet al [12] that if the mortalityratio for smokersto non-smokersfor a

particulardisease is k, then factorX must be presentat least k times more
frequentlyamong smokersthan among non-smokers.The Report does refer
to thisresult,withoutcitationofsource,in the sectionon specificity(page 184),
whereits relevanceescapes me. It is clear,however,that Cornfieldet al regard
this resultas one of the key argumentsagainst the constitutionalhypothesis,
in the Summaryto theirpaper [12]:
as theycite it prominently
"The magnitudeof the excesslung cancerriskamong cigarettesmokersis so
greatthat the resultscan not be interpretedas arisingfroman indirectassociation of cigarettesmokingwith some other agent or characteristic,since this
hypotheticalagent would have to be at least as stronglyassociated with lung
cancer as cigaretteuse; no such agent has been foundor suggested."
A simplearithmeticalillustrationof this (mathematical)phenomenonis developed below. Suppose that the population is divided 50:50 into X and Xc.
Suppose that amongstX people the probabilityofbeinga smokeris 0.9 and the
probabilityof gettingcancer is 0.01, and that these probabilitiesare independent. Suppose furtherthat among Xc people the probabilityof being a
smokeris 0.1 and the probabilityofgettingcanceris 0.001, and that theseprobabilitiesare independent.Then the resultingprobabilitiesare in Table 2. The
cancer rate among smokers is 0.00455/(0.00455+0.49545) =0.0091 and the
cancer rate among non-smokers is 0.00095/(0.00095+0.49905) = 0.0019:
is thus 0.0091/0.0019
the mortalityratio forcancer forsmokers/non-smokers
is
=
X
smokers
-4.79. The rate among
0.45/(0.45+0.05) 0.90 and among nonthe
ratio forX forsmokers/non-smokers
is
smokers 0.05/(0.05+0.45) =0.10;
is thus 0.90/0.10= 9.0. This illustrates Cornfield et al's point that the
must be greaterthan the mortalityratio for
X ratio forsmokers/non-smokers
I
that the numericalvalues ofthe parameters
not
do
feel
smokers/non-smokers.
insertedin this simplemodel are ridiculous,yet withindependenceof S and C
forX and forXc we came out witha large mortalityratioforthe populationas
a whole. I do not see whythe factthat the X ratio is largerthan the mortality
ratio is any embarrassmentto the constitutionalhypothesis.
The Report quotes variousresultsfromHammond's latest report[13] which
illustratesvariousphenomenawhichcan be attributedto various constitutional
hypotheses,thoughone could also easily constructothercausal hypothesesin
each case. The "effect"of the use of tranquilizers,for example is shown in
Table 3. (All numbersquoted in Tables 3-6 are age standardizeddeath rates
fromall causes formales.) Recallingthat this data was collectedby volunteers,

736

AMERICAN

STATISTICAL

ASSOCIATION

JOURNAL,

SEPTEMBER

1965

TABLE 3
Cigarettes
20 + a day

smoked

.Never
Tranquilizes
TranqullCilzersregularly

1501
2286

755
1308

Do not use
Use

TABLE 4
Fried Food

Never smoked
regularly

Cigarettes
20 + a day

No Fried Food Eaten

Fried Food 3-4 timesa wk.

1208
642

2573
1714

it is probablethat some tranquilizerusers did

oftenin a "social" environment,
not admit to this habit, and thereforeappear in the "do not use" row, hence
raisingthe death rate forthisrow.The real "effect"oftranquilizersis therefore
probablyeven moredramaticthan the table shows.
Tables 4, 5, 6 are fromHammond; the firsttwo are not quoted in the Report.
Hammond's data also show the well known,but as far as I know,never satisin death ratesbetweenmarried,single,widowed,
factorilyexplained,differences
and divorced persons. It also shows a marked hereditaryeffect,a marked
heighteffect,and a moderateeducation effect.Hammond does not give crossotherthan on smoking.
classifications,
and
It is noteworthythat in all the above two-way cross-classifications,
othersin Hammond's reportnot quoted here,the smokingcategoryalways has
TABLE 5
Sleep
<5 hrs.
5 hrs.
6 hrs.
7 hrs.
8 hrs.
9 hrs.
10+hrs.

Never smoked
regularly

Cigarettes
20 + a day

2029
1121
805
626
813
967
1898

3936
2655
1601
1426
1562
1729
2694

TABLE 6
Exercise
None
Slight
Moderate
Heavy

Never smoked
regularly

Cigarettes
20+ a day

834
579
486
474

1416
1347
1065
998

A REVIEW

OF "(SMOKING

AND HEALTH"

737

the higherdeath rate. The Report mentionsan unpublishedanalysis supplied

to the Committeeby Ipsen and Pfaelzer,withthe conclusion,"Further,the correlationofany ofthesevariableswithcigarettesmokingwas too weak to reduce
markedlythe correlationof cigarettesmokingwithmortalityafteradjustment
forthe othervariable." It is a pity that the Report does not give any detail
on this analysis,as it miaybe that thisanalysiswould have carriedthe greatest
conviction.However,this is speculation,and we can onlyjudge the Report by
what it containsand by its referencesto the publishedliterature.
One can interpretthe resultsof Tables 3-6 in various ways. The association
with longevityof parents and grandparentsprobably has a geneticinterpretation,thoughit is possibleto interpretit environmentally:
it could be that the
grandparentssettled in an environmentor adopted habits and customs that
werefavorableto longevity,and whichtheypassed on to the later generations.
The association with tranquilizerscould be interpretedas causal, as is the
associationwithcigarettes,but it appears equally plausibleto hypothesizethat
the need fortranquilizers,and/orcigarettes,is merelya reflectionof an underlyingfacet of the human organismsinvolved. Likewise,the associations with
friedfood eating, sleep, and exercisecan be interpretedeitheras causal or as
concealed correlations.
It will be recalledthat the Committee'sassignmentwas "ifpossible,to reach
some definitiveconclusionson the relationshipbetweensmokingand health in
general" (page v).
"This committeewould produceand submitto the Surgeon General a technical reportcontainingevaluationsand conclusions"(page 8).
The Committee's "Judgmentin Brief" was "Cigarette smokingis a health
hazard of sufficientimportancein the United States to warrantappropriate
remedialaction" (page 33).
The detailed conclusionscoverpages 37-40; the most importantare "Cigarette smokingis causally related to lung cancer in men."
"Male cigarette smokershave a higher death rate from coronaryartery
disease than non-smokingmales, but it is not clear that the association has
causal significance."
My opinionis that the Committeehas not establishedthe case forcausality
in lung cancer. My reasonsforthis opinion are, to recapitulate;
(a) This conclusioncan only be justifiedby provingthe genetichypothesis
to be false,and this the Committeehas failed to do.
(b) Even iftherewas no competinghypothesis,the case forcausalityis at the
presenttime significantly
weakened by the grossnonspecificity
of the association and by the absence of hypotheticalphysical-chemicalmechanisms.
My opinionthat the Committeehas failedto offera satisfactoryproofof the
hypothesisofcausalitydoes notimplythat I believethatthe causalityhypothesis has been proved false. It impliesnothingmorethan I believe that it is not
possible to reach definitiveconclusionsat this time.
Although "Recommendationsforactions were not to be part of the Phase
I committee'sresponsibility"(page 8), yet as noted above the Committeedid
concludethat "appropriateremedialaction" was warranted.I take this as an
excuse to append some personalreflections.

738

AMERICAN

STATISTICAL

ASSOCIATION

JOURNAL,

SEPTEMBER

1965

Passing over the questionof whetherany governmentalaction is warranted,

each individual cigarettesmokerhas to decide whetherto make the effortto
give up cigarette smoking. To make this decision he would need to know
(a) the change in his expectationof life assuming that lung cancer alone is
causally connectedwith cigarettesmoking,(b) the change in his expectation
of lifeassumingthat virtuallyall causes of death are causally connectedwith
cagarettesmoking.
The Reportdoes not appear to providethesefigures(searchingin the Report
for any particulartopic is sometimesdifficultas the Report has no index),
whichis rathersurprisingin view ofthe factthat an assessmentofthe magnitude
of the health hazard was one of the Committee's prime responsibilities
(page 8).
The individual cigarette smoker would then have to weight the above
changes in expectation by his assessment of the question of whether the
cigarettesmokingwas a cause. He would have to relate this expected change
in expectationoflifeto the analogous probableexpectedchangesin expectation
of lifecaused by various otherof his activitieswhichprobablycause decreases
in expectation of life, such as over-eating,undersleeping,under-exercising,
travellingby automobile,etc. He would also have to weighthe possibilityof
such adverse consequencesofgivingup smokingas in due course becomingappreciablyoverweight.He would have to assign utilitiesto the expectedchange
in expectationoflifeand to the pleasure and satisfactionhe getsfromcigarette
smoking.
It is unfortunatethat the Report does not give a table of changes in expectationof life,forifthe changeat, say, age 30 was 10 years,then even ifone
assigned a probabilityof 1/5 to the cigarettesmokingcausality hypothesis
beingtrue,the net changeof 2 yearswould give one pause to consider,whereas
if the change at age 30 is only 2 years,then the net change of 0.4 years on an
average expectation of about 40 years mightnot be taken too seriouslyby
some people.
REFERENCES

Zeitscrift
furKrebsforschun[11 Miuller,F. J., "Tabakmissbrauchund Lungencarcinom,"
gen,49 (1939), 57-84.
[2] Doll, R., and Hill, A. B., "A studyofthe aetiologyofcarcinomaofthelung,"British
Medical Journal,2 (1952), 1271-86.
[3] Doll, R., and Hill, A. B., "The mortalityof doctorsin relationto their smoking
habits; a preliminaryreport,"BritishMedical Journal,1 (1954), 1451-5.
[4] Hammond,E. C., and Horn, D., "The relationshipbetweenhumansmokinghabits
and death rates: A follow-upstudy of 187,766men," JournaloftheAmericanMedical Association,155 (1954), 1316-28.
[5] Neyman,J., "Statistics-servantof all sciences,"Science,122 (1955), 401-6.
[61 Berkson,J., "The statisticalstudyof associationbetweensmokingand lung cancer,"
ProceedingsoftheStaffMeetingsoftheMayo Clinic,30 (1955), 319-48.
surveys
[7] Mainland,D., and Herrera,L., "The riskofbiased selectionin forward-going
Journalof ChronicDiseases, 4 (1956), 240-4.
with nonprofessional
interviewers,"
[8] Cutler,Sidney J., "A reviewof the statisticalevidenceon the associationbetween
smokingand lung cancer,"JournaloftheAmericanStatisticalAssociation,50 (1955),
267-82.
[9] Fisher,R. A., Letter to the Editor, BritishMedical Journal,2 (1957), 297-8.

A REVIEW

OF "SMOKING

AND HEALTH"

739

[10] Berkson,J., "Smokingand lung cancer: some observationson two recentreports,"

Journalof the AmericanStatisticalAssociation,53 (1958), 28-38.
[11] Berkson,J., "Smokingand lung cancer,"AmericanStatistician,17 (1963), 15-22.
[12] Cornfield,Jerome,et al., "Smokingand lung cancer: Recent evidenceand a discus22 (1959), 173-203.
sion of some questions,"JournaloftheNational CancerInstitute,
[13] Hammond,E. Cuyler,"Smokingin relationto mortalityand morbidity:Findingsin
first34 monthsof follow-upin a prospectivestudystartedin 1959." StatisticalResearch Section,Medical Affairs,Departmentof the AmericanCancer Society.
[14] Yerushalmy,J., "Statistical considerationsand evaluation of epidemiologicalevidence," in James, George,and Rosenthal,Theodore, Tobaccoand Health. Springfield:Charles C Thomas, 1962, 208-30.
[15] Yerushalmy,J., and Palmer, CarrollE., "On the methodologyof investigationsof
etiologicfactorsin chronicdiseases," Journalof ChronicDiseases, 10 (1959), 27-40.
[16] Eysenck, H. J., Tarrant, M., Woolf, M., and England, L., "Smoking and personality,"BritishMedical Journal,1 (1960), 1456-60.
[17] Seltzer,Carl C., "Morphologicconstitutionand smoking,"JournaloftheAmerican
Medical Association,183 (1963), 639-45.
[18] Dorn, Harold F., and Cutler, Sidney J., "Morbidityfromcancer in the United
States," Public Health MonographNo. 56, 1959.
[19] Fisher,R. A., Letterto the Editor,Nature,136 (1935), 474.
[20] Yule, G. V., "The functionofscientific
methodin scientific
investigation,"Industrial
Fatigue ResearchBoard Report,28 (1924).
[21] Irwin,J. O., "The place ofmathematicsin medicaland biologicalstatistics,"Journal
oftheRoyal StatisticalSociety,Series A, 126 (1963), 1-45.
[22] Fisher,R. A., "Cancer and smoking,"Letterto the Editor,Nature,182 (1958), 596.
[23] Friberg,L., Kaij, L., Dencker,S. J., and Tonsson,E., "Smokinghabitsin monozygotic and dizygotictwins,"BritishMedical Journal,1 (1959), 1090-2.