Sie sind auf Seite 1von 13

Miracles and Statistics: The Casual Assumption of Independence

Author(s): William Kruskal


Source: Journal of the American Statistical Association, Vol. 83, No. 404 (Dec., 1988), pp. 929-
940
Published by: American Statistical Association
Stable URL: http://www.jstor.org/stable/2290117 .
Accessed: 21/02/2011 08:55

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
Miracles and Statistics:The Casual Assumption
of Independence
WILLIAMKRUSKAL*

The primary themeof thisaddressis cautionary:


Statistical
independence is fartoo oftenassumedcasually,without serious
concernforhowcommonis dependenceand howdifficult it can be to achieveindependence (or relatedstructures).
After
initialdiscussion
ofstatistics
andreligion,
theaddressturnsto miracles,especiallyHume'scritique andBabbage'sreply.Stress
is giventheoftentacitor unexamined assumption ofindependence amongwitnesses of a putativemiracle.Othercontextsof
multiple testimony andtheaddressendswithcontemporary
are treated, casualassumptions ofindependence: nuclearreactor
safety, repeatedmeasurements, and so forth.Othertopicsincludeprayer, circularity
of argument, and thetensionbetween
skepticism abouttestimony andthepragmatic needto acceptmostofitprovisionally.
KEY WORDS: Babbage;Hume;Religion;Safety;Testimony.

1. INTRODUCTION givenas an example.Justso, forthosewhotakemiracles


as signsof revelation,the religiousimportof a miracle
FrancisHenryEgerton,Earl of Bridgewater,' died in
1829;in hiswillhe left8,000poundsforthewriting maybe farmoreimportant thanthedailylifein whichit
and
publication ofworkson "thePower,Wisdom,andGood- occurred.In thesameway,therearemisleading apparent
nessof God, as manifested outliersandmisleading falsemiracles,theresults ofpoor,
in Creation."Undereminent
prejudiced,or carelessobservation.If identifiedas such,
auspices,eightclericaland scientific notableseach wrote
theseoutliersandmiraclesareworth littleattention
except
and publisheda treatise.CharlesBabbage,progenitor of
fortheirconnections withthe psychological vagariesof
themoderncomputer, feltthattheeighttreatises didnot
humanobservation orwiththeeffects on behavioroffor-
carrytheirargument farenough,and thatat leastone of
tuitouscoincidences.
themincorrectly arguedthatthepursuitsof scienceare
unfavorable to religion.In addition,Babbagehad views 2. RELIGIONAND STATISTICS
about miracles,viewsin partstatistical. So in 1837 he
publishedwhathe calledTheNinthBridgewater Treatise: Religionand sciencegenerally are twobasicrealmsof
A Fragment. (A second,revisededitionappearedin1838.) human effort that can hardlyfail to interact andintersect.
I beginwithBabbage'streatment ofmiracles inhisown Gilkey (1982) expresses thatstrongly: "new religious forms
context,thatof Hume's famousargument againsttesti- appear and reappear out of and because of a scientific,
monyformiracles.ThenI discussmiraclesandtestimony technological cultureinresponse,first, to thedemandfor
moregenerally, withspecialemphasison theubiquitous, a credible system of symbols giving structure, meaning,
andoftentacitassumption
critical, ofindependence among and direction to nature,history, society, and theself,and,
witnesses. I thendiscussovercasualassumptions of inde- second, to the particularsharp anxieties-and eventer-
pendencein othercontexts.If I have a moral,it is this: rors-of a technological age" (p. 71). A current example
Do notmultiply lightly. ofintersection is thedebate about teaching evolution, es-
Miraclesare relevantto statistics in a numberofways. peciallyin secondary schools, and Gilkey's articleis about
One analogyis thatmiracles arelikeapparent outliers that thatdebate.Again,thegreatphysicist Frank(1956)wrote
we studyandworry over(e.g., see Anscombe1978;Krus- of how all scientifictheoriesare relatedto "theirfitness
kal 1960). One mightevensay thatmiraclesare theex- to supportdesirableattitudes inthedomainofethics,pol-
tremeoutliersofnonscientific life.2It is widelyarguedof itics,and religion"(p. 9). Some willbe surprised at so
outliersthatinvestigation of themechanism foroutlying normative a statement. I do notproposetreating further
maybe farmoreimportant thantheoriginalstudythat thebroadthemeof religionand science,and I certainly
led to the outlier;the discovery of penicillium is often willnotexhibit aggressive incompetence bydipping deeply
intotheology.
Now narrowourfocusto connections betweenreligion
* WilliamKruskalis Professor,Department University andthemathematical
of Statistics, sciences,a partofscienceingeneral.
of Chicago,Chicago,IL 60637.Thisis a revision of theauthor's1982 Thereare manysuchconnections without explicitstatis-
addressto theAmerican
presidential Statistical
Association. Thiswork ticalaspects;I thinkof a rangefromastronomical and
was supported by NationalScienceFoundationGrantsSES 8303838,
DMS 8601732,andDMS 8404941.Theauthorthanks thefollowing,who calendarcalculations to recentattempts (Brams1980)to-
providedcriticisms
andsuggestions:KeithBaker,GeorgeBarnard, Nor- wardapplying gametheoryto biblicalstories.
manBradburn, Stephen Brush,WilliamA. Christian,Jr.,JoelE. Cohen, One also finds fatuousor superficial usesofmathemat-
0. D. Duncan,WilliamB. Fairley,R. W. Farebrother, DonaldFiske,
AntonyFlew,JosephGastwirth, BrianGerrish, LangdonGilkey,I. J. ical terminology, forexample,by FrancisHutchesonin
Good, R. M. Grant,BrownGrier,M. J. R. Healy,Paul W. Holland, 1724: "Benevolenceis directly as theMomentof Good
J. Kadane,BenjaminKing,Leslie Kish,Neal Koblitz,SamuelKotz,
NormaKruskal,MartinE. Marty,EdwardOffenbacher, David Owen,
StephenPortnoy,I. R. Savage,EugeneSeneta,StephenM. Stigler, ? 1988AmericanStatisticalAssociation
JudithM. Tanur,RonaldThisted,B. M. Wright, S. L. Zabell, and oftheAmericanStatistical
Journal Association
ArnoldZellner. December1988,Vol.83, No.404,Presidential
Address
929
930 Journalofthe Amercan StatisticalAssociafion,December 1988

andinversely as theirabilities,i.e.-B = MIA" (Thomp- thatmiraclesreallyhappen-yetanotherrelationship be-


son1931,pp. 14-15).Again,DavidHartleywritesin1749 tweenreligionand statistics.
thatW = F2IL, whereF = fearand L = love of God, Amongrelatively recentempirical studiesofprayer, the
and W = loveoftheworld(see Stephen1962,Vol. 2, pp. earliestI knowwas bythegreatFrancisGalton.He rea-
57-58). Suchuses have been roundly criticizedand also sonedthattheroyalfamily andBritish clergymen probably
defendedas usefulmetaphors. In thecase ofJohnCraig, generatedmoreBritishprayersforgood healththandid
whose1699calculation ofthedateoftheSecondComing others,so he comparedlongevity ofmembers oftheroyal
has been frequently derided,a recentbrilliant interpre-family and clergymen withthelongevity ofothers.There
tationbyStigler(1986) providesa sympathetic recasting. are,of course,obviousproblemsofcontroland compar-
In anycase, theargument overapparently superficial use ability,butGaltonpushedahead nonetheless, and came
ofmathematical expressions continues in bitterargument out withnegativeresultsthatscandalizedmanycontem-
overmembership in theU.S. NationalAcademyof Sci- poraries,perhapsas muchforthemethodas theconclu-
ences(see Koblitz1981;Marshall1986). sions.His firstpublication on thetopicwas in 1872,and
Ifwe specializefurther, fromthemathematical sciences heincludedthematerial inhis1883book(see Galton1872,
generally to probability andstatistics,a numberofdevel- 1883). The book also had a moregeneraldiscussionof
opments cometomind.As Pearson(1924)said,De Moivre's whatGaltoncalled theocratic intervention and thepos-
workon the centrallimittheoremwas theological:"he sibility ofascertainingit. In thesecondedition,however,
was determining thefrequency of irregularitiesfromthe thetheometric materialwas omitted, because-said Gal-
OriginalDesignof theDeity.Withoutgrasping thisside ton-it offended thesensibilities ofa numberofreaders.
ofthematter, itis impossible to understand thehistory of I believethathe neverreturned to thetopicin print.34
fromDe Moivrethrough
statistics DerhamandSiissmilch The Galton(1872) articleand othercontemporary ex-
to Quetelet,culminating in the modernprinciple of the changes werereprintedin ThePrayer-Gauge Debate(Means
stabilityof statisticalratios"(p. 404). In anotherplace, 1876).[Galton'spublication wasslightly precededbyTyn-
Pearson(1978) carriedthesequenceoftheologically mo- dall(1871)andHenryThompson, buttheyonlyproposed
tivatedstatisticians
through Florence Nightingale,andadded studieswhereGaltoncarriedone out.] Pearson's(1924-
that"witha modified meaning we mightalmostaddFran- 1930)massivebiography ofGaltondescribes thestudy(see
cis Galton"(p. 286). Indeed,severalchaptersofthelec- Vol. 2, pp. 115-117).Haldane (1928,pp. 243-245)pro-
tures(Pearson1978),especially thosedealingwithThomas vides a slightly distortedaccount(wronglysubstituting
Bayes and RichardPrice,are suffused withtheological children ofclergymen fortheirfathers), andBrush(1974)
motivations and connections. givesa thorough discussion.5 A recentmedicalevaluation
A different themeis theancientreligious fearofbeing of prayerbyJoyceand Weldon(1965) was double-blind
counted,forexamplein a census.Thisappearsinseveral andinconclusive. Another recentstudy(Cullipp1969)gave
partsof the Old Testament;see Madansky(1986). For resultsmoreencouraging to prayer,but its description
generaltreatments of religionand statistics,see Portnoy leavesme withquestions.Of coursethereis a longliter-
(1988) and Bartholomew (1984,1988). atureon thecurative effects offaithinvarioussenses;for
an overviewand entrybibliography, see Frank(1978).
3. PRAYER
4. MIRACLES
I digressto discussstatisticallyshapedstudiesof the
efficacyof prayer.These studiesare interesting per se, We turnnowto miraclesthemselves. Thereis a huge
and theybringus near the topicof miracles.For some literature on miracles,and I have been able onlyto dip
arguethatprayers-atanyrateanswered prayers-arein intoit selectively witha primary interestin evidencefor
a sensemiracles,ifoftensmallmiracles. YetTyndall(1871) miracles,especiallythe validityof testimonial evidence.
wrote-and manyothershave made thepoint-thatno BeginwithDavidHume's1748definition (see Hume1975)
prayer"could call one showerfromheavenor deflect becauseHumewas one ofthefirst6 to applyprobabilistic
towardsus a singlebeam of thesun" withouta "distur- ideasto miracles,and becauseHume'sfamousargument
banceof naturallaw, quiteas seriousas thestoppageof againstmiracles is framed intermsofone probability out-
an eclipse"(p. 36). weighing another.Humedefined a miracleas "a transgres-
Perhapsthefirstrecordedempiricalcomparative trial sionofa lawofnaturebya particular volition oftheDeity,
of prayerwas the competition betweenElijah and the or by the interposition of some invisibleagent"(Hume
manypriestsofBaal (1 Kings18),although theobjectivity1975,p. 115).See Pomeroy(1962,p. 6) forvariantHume
of itsreporting maybe questionedby some.Recall that characterizations of a miracle.
the priestsof Baal builtan altar,put on it a sacrificed Aquinashad earliersaid thata miracleis something
bullock,andcalledon Baal to startthefire.Nothing hap- doneby "divineagencybeyondtheordercommonly ob-
pened.ThenElijahhad a bullockputon hisaltar,poured servedin nature."A morerecentdefinition speaksof a
wateroverit to makethetaskdramatically harder,and "striking interpositionof divinepowerbywhichtheop-
calledon hisdeity.The response wasoverwhelmingly fiery. erationsof theordinary courseof natureare overruled,
At leastone scholar(Dietl1968)hasbuilton thisexample suspended,or modified."See Flew (1967,p. 346) fora
andproposedrandomized trialsas an approachtoshowing highly relevanttreatment.
Kruskal: Miracles and Statistics 931

Everynuanceof definitions like the above has been forremarkable (1975,p. 127). Humegoeson to
events10
examinedandanalyzed.Forexample,theidea say thatexperience,and onlyexperience,
repeatedly informs us of
thata miracleshouldbe striking, remarkable, the Res- the validityof humantestimony and informs us of the
in brightcolor witha fullchorussingingthe constancy
urrection of thelawsof nature.When"thesetwokinds
Messiah,has been bothunderscored and deniedmany of experienceare contrary, we have nothingto do but
times.Hume himself did notsee theneed fora striking substractthe one fromthe other . . . this substraction,
appearance.He said: amountsto an entire
withregardto all popularreligions,
The raising Theraising annihilation;and therefore. . . no human testimonycan
ofa houseorshipintotheairis a visiblemiracle.
of a feather,whenthewindwantseverso littleofa forcerequisite for . . . prove a miracle, and make it a just foundationfor
thatpurpose,is as reala miracle,thoughnotso sensiblewithregardto anysuchsystem ofreligion"(1975,p. 127).
us. (1975,p. 115)
One can tryto formalize thisargument in severalways,
Otherauthorsmightexcludethe raisingof a featherby comparing twoexplanatory hypotheses abouta reported
insistingthatan eventmusthave clearreligioussignifi- miracle,say thechanging ofwaterintowineor thelevi-
cancetobe counteda miracle;forexample,see Swinburne tationofa holyanchorite. Hypothesis H is thatthemiracle
(1970,pp. 7-10). For a discussion in theOld Testament reallyhappened,so the reportis correct,puttingaside
context, see Eichrodt(1967,p. 163).Yetitis hardtoknow questionsof volitionby a deityor otherinvisibleagent.
whatreligioussignificance mightmean.Eliade (1981) in- Hypothesis H is thatthemiracledidnothappenand that
siststhatmiraclesmayappearin themostordinary parts thereporter (perhapsoneself)was mistaken, tricked,or
of everydaylife.He says,forexample,that"Justas I evena liar.A fullertreatment mightconsiderthepossi-
believedin theunrecognizability ofmiracle,so I also be- bility thatthereport wascorrect butmisinterpreted. There
lieved in the necessity. . . of the camouflageof the 'ex- are two samplepoints,R (the miracleis reported)and
ceptional'in the banal, and of the trans-historic in his- R (it is not),to permitan analysis.
toricalevents"(p. 224). I havedifficulty withtheunrecog- A likelihoodapproachmightbe tried,butit seemsnot
nizablemiracle,forhow can thatbe a signor evidence richenoughto includeHume'sargument. Turninsteadto
or explanation?For a fullerstatement of thispointsee a simple,familiar Bayesianstructure; afterall, miracles
Smart(1964,pp. 48-49). usuallyhave a prioriimprobability. (But keep in mind
Therearealsoviewpoints thatsimply takelifeorscience thoseeveryday miraclesstressedbysome.) Supposethen
or manas miraclesin themselves. "The greatestmiracle that0 is the a prioriprobability forH, and 6 = 1 - 0
is theexistenceofthelawsofnature."Or, "All oflifeis thatforH. Let p be theprobability ofcorrectreporting:
a miracle.Whenyoulookat an appleseed,youarereally p = Pr(R I H) = Pr(R I H). Let p- = 1 - p. Then we
lookingat a miniature fruittree.Thisis as biga miracle havea standard2 x 2 table:
as the newsof the Resurrection." These are some quo- R R
tationsfromscientists in The God of Science,a book7of
interviews withscientists aboutreligion(Trinklein 1971, H pO pao 0
pp. 80-81). Prior(1955) says that"Faith . .. is an inward H TOa pO7 0
miracleofGod's mercy"(p. 8). To be moreaccurate,the
statement is madebya Barthian Protestant inPrior'simag- Thefamiliar centralcalculation istowritethecolumn-wise
inarydialogue.Eichrodt(1967)stressesthattheOld Tes- conditional probability as theunconditional one overthe
tamenttakesmiracles in the"widestpossibleterms.Even sum,Pr(H I R) = pOl(pO + jiG). Note thatifp = p- =
thecourseofNatureitselfcountsas a miracle"(pp. 162- 2, thenPr(H I R) = 0 = Pr(H), a naturalresult.Even
163). morenaturally, ifthewitnessis whollytrustworthy (p =
An important problemwithHume'sdefinition is thatof 1), thenPr(H IR) = 1.
explicating theideaof"lawofnature."Another is toshow Thisformulation maybe criticized in manyways.For
thatan allegedmiracleis in factwilledbythedeity,or by example,thepropensity to lie or be dupedis presumably
an appropriate"invisibleagent"-presumably a benign verydifferent if a highlylikelyeventhappensthanif it
one, say an angel,as opposedto a demonor devil(see doesnot;a bridgehandof12heartsis,I think, muchmore
Swinburne 1968,1970,chap.5). likelyto be reported as 13,thanone of13 reported as 12.
Yet I avoidsuchangelictemptations and hastenon to Thusthevalueofp might wellbe takenas different inthe
thecentralquestionof testimony aboutthehappening of tworows.Moreimportant, thereis everyreasontoexpect
a miracle.For testimony, in particular allegedlyindepen- p tobe a function of0; theprobability ofreporting a bridge
denttestimony fromtwoor morewitnesses, formsa cen- handof 12 heartsas 13 is, I surmise,muchgreaterthan
traltopicofthisexposition.89 reporting a handwith4 heartsas one of5 (see Venn1962,
Hume'sdiscussion oftestimony is cynical,
skeptical, and p. 421). These are psychological and historical questions
fullofwit.He concludes thathumantestimony isso subject thatcannotbe neglected.Thereseemsto be remarkably
to bias, passion,self-interest, changeovertime,and so littlediscussion of reasonablenumerical valuesfor0 and
on, that"no testimony foranykindof miraclehas ever p in varioussettings."l
amounted to a probability,muchlessto a proof;andthat, At a morefundamental level,we generally do notset
even supposingit amountedto a proof,it wouldbe op- up such modelsuntilaftera surprising eventhas been
posedbyanotherproof"fromtheweaknessoftestimony reportedor at leastdiscussed.Thusthereis inherent bias
932 Journal of the American Statistical Association, December 1988

ofselectionintheveryactofmodelformulation. Similarly,givethesame answer,the conditional probability is the


thereis somethingarbitrary up thetwo-pointsameas fora singleobserver.
aboutsetting
samplespace. Coins do land on theiredgessometimes, A complication is thatp mightdecreaseas n increases,
andmoreoftenwitnesses havetheirownin-between sum- a phenomenon observableat magicshows.Undersome
maries:neitherwaternorwine,uncertainties abouthow circumstances, the largerthe audiencethe greaterthe
dead therevivedmanwas,andso forth.In somecases,it chanceofmisobservation.
makeslittlesenseto look at theprobability of a simple Babbage wentthrough calculations similarto those
eventrather prob- presented
ofdecreasingly
thanat a tailprobability previously and concludedas follows:
able eventsunderone hypothesis. ... if independentwitnesses can be found,whospeaktruthmore
Humeargued,as I interpret him,that frequentlythanfalsehood,it is ALWAYS possible to assigna numberof
independentwitnesses,theimprobabilityof thefalsehood of whose con-
shall be greaterthan thatof the improbability
curringtestimony of the
Pr(incorrect I R) = Pr(H IR) > Pr(H I R)
observation
miracleitself.
... thewholeofthisargument witnesses.
appliesto independent The
= Pr(correct I R),
observation
of thecollusion,and thedegreeof creditto be assignedto
possibility
witnessesunderanygivencircumstances, dependon factswhichhave
thatis, thatPr(H IR) < . A littlecalculation showsthatnotyetbeensufficientlycollectedtobecomethesubjectofmathematical
thisis equivalent to 0 < -; that is, the unconditional prob-
inquiry.Some of thoseconsiderations whichbear on thispartof the
ability of the miracle is less than the probabilityof subject,thereaderwillfindtreatedofin theworkofDr. ConyersMid-
a lie
dleton,entitled"A FreeInquiry IntotheMiraculous PowersWhichAre
(or mistake, or the like). He argues by anecdote, appeals
Supposedto Have Subsisted intheChristian Church,FromtheEarliest
to experience, and explanation ofhiscynicalviewof hu- AgesThrough SeveralSuccessive Centuries." (Babbage1838,pp. 202-
mannature.[Thatskeptical"2 position wasnotnew.Indeed 203) 13

it longpredatesthe Christian era; see Grant(1952) fora Not surprisingly, otherscriticized"4 Hume alongBab-
detailedhistory of waves of skepticism and belief.]Hume bage-likelines.Muchearlier, Price (1777, p. 418) briefly
might-butI thinkdid not-have quoted the skeptical made the point.Chalmers(1857, pp. 66-69, 130-135)
wordsof Shakespeare,who had Gloucester say "What gave a similarargument beforeBabbage. Chalmers ex-
meansthisnoise?/Fellow, whatmiracle dost thou pro- cusedhis"mathematical styleofreasoning" because itwas
claim?" (Henry theSixth,Second Part, act 2, sc. 2). Still "the best fittedto neutralize the mischievous influence
Humeputtheskepticalcase withspecialforce,at a time superaddedto theskepticism ofHumebythegreatname
of developinguncertainties about Christianity, and in of La Place" (p. 66). Chalmersalso scoldedLaplace on
terms-however earlyand rough-ofprobability. pages114-116and 142-143;I shallreturn to Laplace.
PriceandChalmers arecareful torequireindependence,
5. BABBAGE'SREPLY althoughI do not see properdiscussions of how strong
thatrequirement is. (Indeed,Chalmers sometimes forgets
Hume'sargument wasmetbya stormofprotest, much
andmultiplies toocasually,e.g., p. 122.)In thesameway
too extensiveto reporthere.[One accountis by Brown
themathematician JamesR. Youngcarefully assumesin-
(1984,pp. 89-91).] But I do wantto describeBabbage's ex-
dependence-nocollusion,in hislanguage-without
answerto Hume.Babbagesaid,ineffect, "David Hume,
aminingthestrength of thatassumption. Young'scalcu-
youmaybe rightfora singleobserverthat0 is less than
lationsare muchlikeBabbage's;a firstclaimby Young
p5,but supposethatthereare severalobservers, and in-
(1844,pp. 366-367; 1846,p. 80) is thatBabbage (1838,
dependent ones,whoagreein reporting themiracle,that
pp. 198-199)madea smallerrorin calculation thatdoes
is,inR. Isn'titobviousthat,foranyfixed0 > 0 no matter
affecthis mainconclusion.In fact,it seemsto me that
howsmall,we can findan integern suchthatagreement
YoungmisreadBabbageand thatthetwoagree.
ofn independent observers onR willproducea conditional
Young'ssecondclaimedadvanceis moreinteresting. He
probability forH thatis arbitrarily close to 1?" Wellof
points out thatboth Hume's and Babbage's treatments
coursethat'sright in an algebraicsenseprovidedthatp >
referto a specifiedmiracle;in fact,initialreportsofmir-
p, thatis, a
that correct observation ismoreprobablethan
acles referto previously unspecified miracles."Suppose,
an incorrect one.
forinstance,"saysYoung(1844), thata persontestifies
Indeed,theconditional probability ofH, giventhatall
to thefactthathe saw a dead manraisedto life;andthat
n independent observersagreein reporting thatit hap-
tenotherpersonswitheverydisposition to deceive,but
pened,is withoutcollusion,testify to the same thing.Now, sup-
pft 1 posingthatthesetenpersonswerelimited withinthevery
narrow rangeofonlytenfabrications suitabletotheirpur-
~P~0 posesofdeception:theprobability thattheywouldall fix
upon the particular miraclementioned is 1/(10)1o''(pp.
wherewe assume0 < 0 < 1. Ifp >75, thenthisprobability363-365).Thisdevelopment is an earlydiscussion ofthe
as a function ofn is monotone increasing tothelimitunity. familiar problemof selection.It is also an at-
statistical
Notethatthisassumesthatp is thesameforall observers temptto get aroundthe Hume-Babbagerequirement p
and nota function ofn. It also assumescrucially thatthe >2. But of coursetheprimary weaknessin Young'sar-
observers behaveindependently. Attheoppositeextreme, gument is thedifficulty in establishing theabsenceofcol-
if the observersbehavein a whollydependentwayand lusion.
Kruskal:Miracles and Statistics 933

Thatsomeprobability is verysmallsayslittleornothing utor.In mostof thisworktherewas littlereference to


byitself,or else everybridgehandwouldbe astonishing.empiricalevidenceand greatrelianceon a priorisimpli-
One naturalstatistical way of formalizing Young's ap- fying assumptions: independence, equalprobabilities over
proachis to considerthe alternative that,if one of the people,and probabilities of errornotdependingon cir-
witnessesin factsaw a dead manraisedto life,he would cumstances. Therewaseventually a revulsionagainstwhat
(no matterhowmucha liar) reportit withprobability qV John StuartMill called the "opprobrium of mathematics"
> .1. That leads to a likelihoodratio of (V/.1)10 for the (Mill1952,book3, chap.18,sec. 3, p. 353; 1973,p. 538),
samplepointin whichall reportthe revival-assuming an expression translated intoFrenchbyBertrand(1888,
independence. p. 327) as "la scandaledes Mathematiques." I hastento
The literature on miraclesincludesmuchon qualitative add thattherewas also workof theera thattriedto be
consistencyamongwitnesses, forexample,amongtheNew morerealisticand thatdealtcarefully withempirical evi-
Testament Gospels.A widelyreadnineteenth-century ex- dence;in particular I citeCournot(1838) and Bienayme
ampleis TheLifeofJesusbyStrauss(1892,e.g., see pp. (1838).Forfurther discussion see HeydeandSeneta(1977,
742-743). ErnestDeWittBurton,president of the Uni- especiallysec. 2.4, pp. 28-34).17
versityofChicago,devotedmuchofhisscholarly energies Therewas, in anycase, a turning againstdirectprob-
to reconciling,and (as he putit) findingharmony among, abilisticanalysesof matterslegaland legislative; onlyin
the Gospels(e.g., see Stevensand Burton1905). More recentyearshas suchwork-done farmorecarefully-
recentrelatedworksareMcArthur (1965),whichdiscusses had a rebirth.Yet even todayegregiousoversimplifica-
possibledependenciesamongthe Gospels,15 and Perrin tionsare madeinlegalsettings. A popularexampleis the
(1977). notorious Collinscase, in whichmultiplication of alleged
probabilities ofcharacteristics wascasuallydone;see Fair-
6. OTHERFUNDAMENTAL ISSUES ley and Mosteller(1974) forcitationsand a tolerantex-
In seriousstatisticalthought abouttestimony, as about position.An extensivediscussionin thelegal literature,
is McCormick (1972,pp. 491-499).
mostotherapplications, fundamental problems mayques- withmanyreferences,
tiontherelevanceofprobability ideas at all, or pointout See also theexposition by Eggleston (1978),inwhichthe
ambiguities in framing a model.As Venn(1962) putit, assumption of independence is carefully stated,butas a
"the individual presentedhimself, and thetaskwas im- kind of incantation without due attention to itsstrength
posed uponus of selecting therequisitegroupsor series and the difficultyof confirming it.18
to whichto referhim. . . . we have to selecttheconditions
of frequency out of a pluralityof moreor less suitable 8. RECAPITULATION
ones" (p. 395). In effect,Vennshowshowp maywell
I pause to recapitulate. Starting withquestionsabout
dependon thenatureoftheeventin question,thuschal-
testimony formiracles, we haveseenthatonecentralissue
lengingourabilityto setup a usableprobability spaceat
all. The pointhad been raisedearlier,forexampleby is thatofindependence-or allegedindependence-among
witnesses.We havetracedbriefly thehistorical develop-
Chalmers (1857,pp. 140-141)as he chidesHumeforshift-
ment of modelsfortestimony generally, including testi-
ingreference populations.
mony in circumstances.
judicial Beforeturning tocontem-
So thereference setmaybe questioned,and thereare
further For example,in surveys poraryexamples,one maywell ask whytheassumption
complications. withsev-
ofindependence is so widespread.
eral or manyinterviewers it maybe sensibleto regard
One answeris ignorance.Many otherwiseestimable
themas randomly chosen,thusintroducing dependence
peoplewillwritethat,forexample,"The combination of
amongtherespondents interviewed bythesameenumer-
thesetwoprobabilities-that thecorewilldryoutunder
ator.In othercircumstances, especiallywithone or two
accidentconditions andthatsafeguard systems willfailto
interviewers only,wemaydecidetoregardanyinterviewer
refloodthe core-is the productof the two individual
effectas fixedso thatthereis no stochastic dependence. or 1 in 1 million"(Lapp 1974,p. 72).
probabilities,
Thereis a similarambiguity insomediscussions ofcovari-
Far moreimportant thansimpleignorance, inmyopin-
anceanalysiswherenearbyplotsofgroundin agricultural
ion,is seductivesimplicity: It is so easyto multiply mar-
experimentation aretakento havecorrelated fertilities
on
ginalprobabilities,formulas simplify, andmanipulation is
one page and consideredas bearingunknownfixeddif-
relativelysmooth, so the investigator neglectsdepen-
ferentialfertilities
on the next.I thinkthereare episte-
dence,orhopesthatitmakeslittledifference. Sometimes
mologicalissues16 herethatremainto be sortedout.
thathopeisrealized,butmoreoftendependence canmake
7. FRANCE a tremendous difference.
Thereareintermediate positions intheconstant tension
Beforecomingto contemporary examples,I turnback betweenverisimilitude and simplicity.19A grandthemein
toFrenchthought. StartingwithCondorcet andLaplace- thedevelopment ofstochastic modelsis theassumption of
especially inthelatter'sfamousPhilosophical Essay-cal- conditional independence, withthecondition oftenturn-
culationssomething likethosesketchedearlierfortesti- ingon unobserved, hypothetical variables.Twoexamples
monyto miracles,decisionsaboutlaw-court cases, and are factoranalysisand latentstructure theory.
deliberations oflegislatures.Poissonwasa majorcontrib- In myviewthereis a centralobligation to facesquarely
934 Journalofthe American StatisticalAssociation,December 1988

whatwe do and do notknow,and to studyrobustness of Titanic.BeforetheTitanic hitan icebergandsankin 1912,


conclusion againstmistakes in a prioriassumptions ofin- shewasdeclaredunsinkable becauseofherdouble-bottom
dependence,conditional or not.20 hullandwatertight compartments. Unfortunately theice-
bergcut a 300-footgash on one side thatfloodedfive
adjacentcompartments. In addition,the radiooperator
9. CONTEMPORARY EXAMPLES on a nearbyshipwas asleep,the Titanicwas goingtoo
Over-easyacceptanceof independence, and strugglesfast,and therewereinsufficient lifeboats.
withthecomplexities of notassuming it, continueabout It is ironicthatMacholtooktheothersidein a debate
us. I givea fewexamplesbeyondthosealreadymentioned, withthelateJ.Kieferin an entirely differentdomain,the
starting withaccidents. taxonomy and classification of mushrooms. Machol,to-
Estimatesof accidentprobabilities in nuclearreactors getherwithR. Singer,had strongly suggestedassigning
are important, and muchin thenews.The difficulties of mushroom speciesto generain accordwithpseudolikeli-
makingthoseestimates, as withotherlow-probability ac- hood ratiosformedbymultiplying marginal probabilities
cidents,turnon the relatively fewseriousaccidentsob- ofcharacteristics (likehavingamyloidsporesorhavingan
servedand on uncertainties in setting up physicalmodels intermediate lamellaewidth).The procedure, whichwas
foreventsleadingto accidents. The temptation to assume called by Healy (1979,p. 169) in the medicaldiagnosis
independence at variouspointsmustbe strong;otherwise context"idiot'sBayes," is in factthe correctBayes as-
howcan one geton withthejob? signment procedure ifall indicators are assumedindepen-
Mostaccidentanalysts understand, ofcourse,thatone dent. Healy pointsout thatthe idiotBayes procedure
cannotsafelymultiply marginal probabilities. For exam- sometimes workssurprisingly well,butKiefer(1979)vig-
ple, afterexplaining multiplication underindependence, orouslyarguesthatresponsibility forshowing thatitworks
Bethe(1976)saysthatone cannotalwaysdo that:"There or notis theproposer'sratherthanthereader's.Kiefer's
can be 'commonmode'failureswhereone eventtriggers analytical and criticalarticleappearedin themycological
twoor threefailuresofessentialelementsofthereactor; literature; it containsa fullbibliography. Macholreplied
in thatcase the probability is the same as thatof the in 1980.
triggering event,and one does notgetanybenefitfrom In theearlierdiscussion ofprayer,5 I mentioned natural
themultiplication ofsmallprobability numbers. Theprob- continued interest bytheBishopsof theChurchof Eng-
ability ofsuchcommon-mode failures is ofcoursethemost land. A recentepiscopalstatement aboutreligiousbelief
difficult to estimate"(p. 25). Exactlyso. In factmany (House of Bishops1986)deals in considerable partwith
accidentpathways areinbetweenindependence andcom- miraclesand testimony forthem.It is explicitaboutde-
pletedependence.One facesthedifficult choicebetween pendenceand independence, but its usage is, I think,
simplifying assumptions-the simplest is independence- somewhatvague.Thus,forexample,it saysthat"if we
and inserting guessesabout manycomplexconditional bear in mindthe dependenceof Matthewand Luke on
probabilitieswheredirectempiricalevidenceis non- Mark,thereare onlytwo independent witnessesin the
existent ortenuous.Or arbitrary averagesofextreme pos- NewTestament totheinstitution oftheEucharist, namely
sibilities maybe taken;fora criticalanalysisof thatap- Markand Paul" (p. 28). WhyMarkand Paul are taken
proach,see Lewis(1980,p. 61). as independent is not made clear. The same vagueness
It is notdifficult to finda varietyof statements at dif- maybe foundin somehistorical literature,forexample,
ferent levelsofcare.Deffeyes andMacGregor(1980),for Garraghan(1946, pp. 294-295 and elsewhere)as cited
example,givetheclassically ingenuous form:"theprob- somewhat misleadingly byBurns(1981,p. 296). Bothof
abilitythatan ore depositwillbe formed. .. is determined thesebooks containmuchmaterialrelevantto our dis-
bymultiplying theprobabilities thateach essentialingre- cussion.
dientwaspresent"(p. 68). Swerdlow (1979,pp. 529-530) We see thensomeof thetrapsin gratuitous or casual
takesa contemporary historian of astronomy to taskfor assumptions of independence. A grandironyis thatin-
naivelymultiplying probabilities. dependence seemsrareinnature,andwhenwereallywant
Lorenz(1974) recognizesthe need forindependence, it,we go to greatpainsto achieveit,forexamplein the
butpasseslightly overjudgingit: "The improbability of production ofpseudorandom numbers forselective
service
coincidental similarity is proportional to the numberof lotteries(see Fienberg1971), and in randomization for
independent traitsof similarity, and is, forn suchchar- allocationoftreatments in experiments. (Thesetwocases
acters,equal to 2n-1'' (p. 230). He refersto striking re- are not,strictly speaking,aimedat fullindependence be-
semblances in shapeof,forexample,a shark,a dolphin, cause of theirwithout-replacement structure;thecentral
and a torpedo.Otherexamplesmayreadilybe foundin pointis thattheyareaimedat jointdistributions ofknown
bothordinary scienceand exoticareas,forexample,es- formforthesamplepoint.In factthoseknownformsare
timating theprobablity ofextraterrestrial lifeor theprob- usuallycloselyrelatedto randomsampling.)
ability ofa worldwarnextyear.21 Estimating probabilities In mostreal cases thereis noticeabledependencebe-
likethelattertwostretches to thelimit-perhapsbeyond tweenphenomena.We are surprised, forexample,when
it-the appropriateness ofprobabilistic ideas. socialpsychologists working withsentiments of content-
How important dependencecan be is vividly described ment(see Bradburn1969,pp. 57-61) findlittleor no
byMachol(1975)inconnection withthesinking oftheSS association betweenputatively polartraitsofpositive hap-
Kruskal: Miracles and Statistics 935

pinessand negativediscontent.Similarly(Spence and holeinreceivednaturallaw,andarguesmoreorlessquantitatively (pp.


Helmreich 1978)putatively polartraitsofstereotypic mas- 265-266) againstGalton's empirical analysis. Partof thatanalysisshows
thatclergyliveabouttwoyearslongerthanlawyersand doctors;Ro-
culinity(self-confidence and competitiveness) versusste- manessaysthatonlya fraction of prayers byclergyare trulysincere,
reotypicfemininity (gentlenessand concernforothers) sayhalf,thatonlya fraction ofthoseareforclergy, sayhalfagain,and
turnout to be almostvanishingly correlatedgivensex. thatonlyhalfoftheremaining prayers areforlengthoflifeas opposed
[Cautionary comments:(a) weak correlation is not the tootherboons.Multiplication givesaneightfold dilution,andmultiplying
sameas independence; (b) theaveragesdiffer in standard again two years by eight provides an effective advantage of 16 years!
Romanes,to be sure,recognizes thegapsin thiscounterargument and
stereotyped ways.] saysthathe givesit to show"thatthestatistical methodis nota trust-
An almostuniversal assumption instatisticalmodelsfor worthy instrument withwhichto gaugethephysical efficacy
ofPrayer"
repeatedmeasurements of real-world quantitiesis that (p. 266). Romanes'sthinking aboutreligion fluctuated,probably under
thosemeasurements are independent,22 yetwe knowthat theinfluence of CharlesDarwin.An accountof perhapsquestionable
suchindependence Berkson,Magath,andHurn accuracyis givenbyRomanes'swidow(1908).
is fragile.
4. The omission, according to Pearson(1924-1930,Vol. 3B), was at
(1939), forexample,showedhow laboratory blood-cell "theurgentrequestof thepublishers"(p. 448). In hisprefaceto the
countspresentsurprising correlations.Repeated mea- secondedition,Galtondescribestheomissionambiguously, butplainly
surements bythesameobserverare boundto be affected saysthathe finds"nothingto recant"in theomittedchapters.
bymemory ofpriorobservedvalues.To avoidthatprob- 5. Therewas a use of theprayer-gauge idea by Stead(1891) in his
lemsomemeasuring deviceshavearrangements forblind about
article a then-current scandal centering on AlbertEdward,Prince
ofWales,andallegedcheating at baccarat.Steadcalculated thatat least
haphazardresetting ofthezeropointbetweensuccessive a thousandmillionprayers forthePrinceofWaleshadbeenmadeduring
observations.23Mahalanobis(1947)suggested carryingthis hislife,onlyto havehimbecometheobjectofa virulent pressattack
furtherbyusingdifferent scalesforseparateobservations,overcardcheating.Benson(1930,pp. 211-213)givesa lucidaccountof
each withvariantscale markings-perhaps even nonlin- theincident, buthe putsSteadwiththeattackers whenin factSteadis
earlyrelated!Thereis a relatedsubstantial literature on sympathetic toward the Prince-if not to the effectivenessof prayer.
Steadexplains thePrince'ssuperficial lifeas theconsequence ofboredom
processcontrol,in the sense of independent identicallyandlackofseriousresponsibilities. Turner(1974,pp. 59-63) describes
distributedobservations. Muchofthatliterature dealswith anothercontroversy focusedon thePrinceof Walesand prayer-this
in expectedvalue,butsomediscussesdependence. oneaboutseriousillnessandhisrecovery.
shifts Turner's article
provides valu-
Page (1978) givesan excellentoverview. able furtherdiscussion abouttheargument overefficacy ofprayer.
to Inrecent yearstherehaveagainbeenBritish commentaries onpossible
So we havecomefrommiracles, through testimony, between
relationships prayerandhealing, commentaries published under
inflatedorinsidious claimsofindependence andhowthey eminentand established auspices(see Archbishops' Commission 1958;
canbe damaging. I beganwithCharlesBabbageandI end BritishMedicalAssociation1956;Ramsey1971).I havenotedno ref-
by quotingWilliamSchwenkGilbert,anotherwise En- erenceinthesepublications toChalmers's explanation,Galton'sanalysis,
glishman. In Act 2 of The Yeomanof theGuardFairfax ortheprayer-gauge debateotherwise. Therearecriticisms-mildly stated
asks,"And thoudidstsee all this?"Pointanswers, yetperhapsdeep-of the BritishMedicalAssociationreportby the
Archbishops' Commission in(especially thelatter'schapter3).
Debate aboutmiraclescontinues withintheChurchof England.In
Aye,withbotheyesat once-this and that.The testimony of one eye
is naught-hemaylie. But whenit is corroborated particular,I citethe Church's (1986) Reportof theChurchofEngland
bytheother,it is
goodevidencethatnonemaygainsay.Here are bothpresentin court, Bishops;itaroseafterYorkMinster wassetafirebylightning soonafter
readyto swearto him. the consecration thereof now-Bishop David Jenkins, who had been
outspokenly skepticalaboutthefundamental Christian miracles.Two
descriptivestatements arebyThomas(1985)andKramer(1985,pp. 80,
NOTES 86-89). A strongly pro-miracle viewpoint is givenbyBerry(1986)to-
gether withmanyfurther references. Inthecontext ofdependence among
1. He wastheeighthandlastEarl,a renowned eccentricwithfamily witnesses, I shallreturn laterto a 1986statement bytheHouseofBish-
money,largely fromthebuilding andoperation ofEnglishcanals.Falk ops.
(1942)givesa colorful account.The Earl'stestamentary intentionswere 6. Notliterallythefirst.BrownGrier(Northern IllinoisUniversity)
carriedoutbythePresident oftheRoyalSociety,assistedbytheArch- has collectedotherexamples,of whichan unusually interestingone is
bishopof Canterbury and theBishopof London.See Babbage(1838, theworkofHooper[1689(especially chap.4), 1699]on transmission of
p. xxi)andBarnes(1965,p. 857). humantestimony. The centrality and persistence of Hume'sargument
2. A philosophicaldiscussion of therelationshipbetweenapparent nonetheless makesit a goodpointofdeparture. As Stephensays"The
miraclesand "irregular or discrepant fact[s]"generallyis givenby factthattheargument isbeinganswered tothisdayprovesthatitsefficacy
McKinnon(1967,pp. 312-313). is notexhausted.Everynewassaultis a tacitadmission thatprevious
toregardanswered
3. It maybe an oversimplification prayers as small assaultshavenotdemolished thehostileworks"(Stephen1962,vol. 1,
miracles. Forexample,Chalmers (1842)has an ingeniousdistinctionto chap.6, sec. 1, p. 262).
explaintheapparent inconsistency betweentheefficacyofprayer (taken Burns(1981) treatspredecessors of Hume;see pages10, 59-60, 75
forgranted)and theobserveduniformity of nature,save forthefew (about Annet),77 (Woolston),89 (Wollaston),and 90-95. [Annet's
"wellaccredited miraclesoftheJewish andChristiandispensations" (p. treatment (1744,especially p. 64) is striking,although itlackseventhe
237). Chalmers pointedoutthatwe areunableto tracecausativechains roughprobability idea thatHumeuses.Annetstresses theshrewd point
innaturebackformorethana fewsteps,andthatprayers areanswered that,if a miracleworkerreallyhad thepowerto changenaturemar-
byunobservable divineinterventionfurther
backinthechain.Real mir- velously, he might ipsofacto,andmuchmoresimply, deceivethesenses
acles-thoseexceedingly rareevents-arisefromobvious,visibledivine ofwitnesses.] Burnsgivesa systematic outlineofHume'sseveralargu-
intervention.The samedistinction had beenmadebyChalmers earlier mentsaboutmiracles;I discussprimarily theargument basedon testi-
(1857,pp. 149-150).He didnothesitate torepeata favoritetheme,and monyand probability, calledbyBurnsthea prioriepistemological ar-
I notefurther thathe wrotethefirst Bridgewater Treatise! gument.
Romanes's(1874)bookstresses betweenanswered
thesimilarity pray- 7. Marty(1977)callstheTrinklein book"obscureandalmostnaive."
ers and miracles(especiallypp. x, 174-187),treatsconstructively the Trinklein, saysMarty,approacheda "ratherrandomsampleof presti-
dilemmathata propermiracleor an answeredprayerseemsto teara gious. .. scientists . . .. The opinions he heardwereas variedas those
936 Journalofthe American StatisticalAssociation,December 1988

. . . from38 randomly selectedpatronsof anyfriendly neighborhood dispensewiththeauthority ofexpertsand,in manycases,dependence


tavern"(p. 31). on menofhigher intelligence andlargerexperience thanourownis the
8. Beforegoingon, however, it is worthmentioning againthepos- mostreasonablemeanswe canuse oftheattainment oftruth"(p. 638).
sibilityofextraordinary eventsthatarenotmiracles; thismaybe called, Howeversensitive wearetothevagaries oftestimony, wesimply must
following Hume,theIndianPrincequestion.A SouthIndianPrince, acceptmostofit,atleastprovisionally. Wehaveneither timenorenergy
saysHume(1975),is toldthatwaterbecomeshardandsolidwhenthe to checkmorethana littleofwhatwe havebeentoldor read.Thatis
temperature goesdownenough.Theparochial Prince,continues Hume, certainly trueofdailylife.In scientific life,it is impossible foreachof
oughttobe skeptical andrequire strong testimony, buttheallegedaqueous ustorecapitulate morethana tinyfraction ofotherpeoples'observations.
behavior is "not contraryto his experience . . . [but rather]not con- Eveninmathematics thereis an analog:Themostenergetic andstringent
formable to it" (pp. 113-114).He simply had neverobservedwaterat mathematician cannotcheckandconfirm morethana smallfraction of
lowtemperatures. TheexamplegoesbackatleasttoJohnLockein 1690 statements andproofsinthemathematical literature. As Millsays,"Au-
(see Locke1959,chap.9, sec.5) interms oftheKingofSiam.Something thority. . . is the evidence on whicheven the wisestreceive all those
likethatis thestoryofmeteorites, whoseexistence was at first denied truths of science,or factsin history or in life,of whichtheyhavenot
foryears(see Westrum 1978).Thereis an apocryphal Hume-like an- personally examinedtheproofs"(Mill 1969,p. 407). Thusprofound,
ecdoteaboutThomasJefferson: Wheninformed thattwoYaleprofessors persistent skepticism wouldmakelifeimpossible. I findthatparadoxical
hadreported a meteorite fallin 1807,Jefferson is saidtohaveremarked andpuzzling.
that"It is easierto believethattwoYankeeprofessors wouldlie than 13. Humein factrefers to multiple witnesses at severalplacesin his
thatstoneswouldfallfromheaven"(Mason1967,p. 429). Whether ball essay(e.g., Hume1975,pp. 112,116,121)butthereferences are brief
lightning reallyexistshasbeenquestioned foryears(Garfield 1977;Stew- and in passing.Swinburne (1968,1970)picksup themultiple-witness
art 1878,p. 403; Uman 1982,p. 535). The Stewartdiscussion deals pointand wonderswhether Humehas overstringent standards of evi-
explicitly withsupposedly independent observers. Of coursetheseex- dence.Middleton's bookwas widelyinfluential; it is discussed, forex-
amplescan be morethanmatched bymanycasesoffalse,tendentious, ample,by Stephen(1962,Vol. 1, especiallypp. 222-230). I do not,
hallucinatory, and so forthobservations of apparently extraordinaryhowever, findthequestionofindependence addressedbyMiddleton.
events.Mostof thesewouldnotbe calledevenpotentialmiraclesby 14. Therewerealso restatements of Hume'sargument, notablyby
mostmiracleenthusiasts, primarily becausethereis littlereligious rel- Mill(1885,pp. 217-241;1969,pp. 470-489),whoincludesa bitter pas-
evancein ice, meteorites, balllightning, andthelike. sage(pp.236-237;479-480)ontheweakness oftestimony forthebiblical
9. A basicdivergence intheliterature on miracles corresponds to the miracles. Venn(1962,chap.17) tooka dimviewofprobabilistic calcu-
age-olddebateaboutfreewillanddivineintervention. To putitcrudely: lationsabouttheveracity of jointtestimony. A majorreasonforhis
Did thecreativedivinity starttheuniverse in a granddesignand then skepticism wasthat"practically ourmainsourceoferrorandsuspicion
simply sitbackandwatchit spin?Or does divinity tinker andinterfere is thepossibleexistence ofsomekindofcollusion.Sincewe canseldom
fromtimeto timebysuddenly introducing miracles? entirelygetridofthedanger, andwhenitexistsitcanneverbe submitted
Babbage(1838,pp. 32-49, 168-171)seemsto takethefirst view;he tonumerical calculation, itappearsto methatcombination oftestimony
regardsmiraclesas preordained apparentshifts in naturallaw,butnot ... is yetmoreunfitted forconsideration in Probability thaneventhat
realshifts-rather, partsofa moregenerallaw.His analogis to a com- ofsingletestimony" (p. 428).
puterprogram thatgivesthesequenceofnatural numbers fora thousand JeanJacquesRousseauearlierpresented a skeptical approachto such
yearsandthensuddenly shifts toanother sequenceinobediencetoinitial calculations, presumably notintermsofexplicit probability calculus.In
programming. Millarguesthattheencompassing law-of-nature concept "The Creedofa SavoyardPriest,"he asked:"Whowillventure to tell
does notgivea reasonableinterpretation formiracles(Mill 1969,pp. me howmanyeye-witnesses are requiredto makea miraclecredible?
473-478).Millsaysthata miracle "mustbe produced bya direct volition, Whatuse areyourmiracles, performed inproofofyourdoctrine, ifthey
withoutthe use of means;or at least,of anymeanswhichif simply themselves requireso muchproof?"(Rousseau1911,p. 263).
repeatedwouldproduceit" (pp. 224,474). A nineteenth-century eloquentsupporter ofthebiblicalmiracles and
Some thinkers seemto takebothviewpoints at differentplaces.An theircentral importance forChristianity wasRichardTrench.It is wryly
exampleis Maimonides; see Hartman(1976,pp. 157,251) and Heller ironicthathisviewsaboutmultiple witnesses shouldundercut thear-
(1958).In Maimonides's (1963)GuideofthePerplexed thegenerallaw guments putforward byothercritics ofHume.Trench(1860)wrotethat
viewpoint is takenon pages480-484,butthemoretraditional interven- "howevermanythey[thewitnesses] maybe, theyare alwaysbuta few
tionistposition appearsonpages178,498-499,and528-529.Fortheolo- compared withthemultitudes whoattesta factwhichexcludestheirfact,
gianswhoregardcreation oftheuniverse as timeless, oroutsideoftime, namely,theuninterrupted succession of a naturalorderin theworld"
theproblem disappears, butfresh problems maythenarise.Forexample, (p. 61). Trench'sdiscernment is in anycase questionable. He asserts,
howcan one reconcile a timeless universe andthespecifically historical forexample,thatpeopleare lighter whenawakethanwhenasleep,as
character ofChristianity? firstobservedbyPlinyand verified by"everynursethathas carrieda
10. Humeusedtheconceptofprobability in a rather roughway.For child"(p. 232).Another curiousTrenchposition is suspicion aboutvari-
discussions ofthestateofprobability atorbefore HumeseeBurns(1981), ationamongauthoritative accounts ofhistorical happenings, exactly the
Van Leeuwen(1963),and Peirce(1958,pp. 305-306).See also Sobel oppositeviewtothatexpressed bymostdefenders ofthebiblicalmiracles
(1987) and Owen (1987) forrelevantmaterial, especially on Bayesian (thatvariation in accountsis natural,human,andincreasescredibility;
interpretations ofHume'sstatements. Patey(1984)describes thedevel- p. 233).
opmentofprobability concepts. 15. Thereis a relatedliterature aboutsurprising degreesofagreement
11. Thereis a sparsebutextended literature presenting algebraic for- amongtheFathersoftheChurch(see especially Daille 1843).Daille in
mulasfortruthand testimony undervariousconditions. An excellent effect discussesdependence amongtheviewsoftheFathers,andeven
criticalsummary is byZabell (in press). drawsanalogiesto contagion ofillness.In thelastchapterofhisbook,
12. One themeofthisessayhas beena skepticism abouttestimony; however, he seemsto go theotherwayandarguesthatnearunanimity
indeed,statisticians-along withlawyers, historians, andothers-byin- ofopinionstrengthens thecase (see pp. 144-145,170,268-270,349).
clinationand training listenwithspecialcautionto humantestimony. 16. Allegedcircularity ofargument isanother basicquestion thatarises
YetI mustcomment inpassingthatall ofus takemosttestimony at face in variousrelatedways.Perhapsthemostfundamental wayis thevali-
valuemostofthetime.We couldnototherwise carryon ourlives. dationoffaithvia miracles described in a HolyBook; theBookin turn
Locke(1959,book4, chap.15,par.6) pointedoutin1690thedangers is validatedby the faith.As MatthewTindal,an earlyEnglishdeist,
inaccepting opinions ofothers:"sincethereis muchmorefalsehood and wrotein 1730,"It's an odd jumbleto provethetruth ofa bookbythe
erroramongmen,thantruthand knowledge . . . menhavereasonto truth ofthedoctrines itcontains, andatthesametimetoconcludethose
be Heathensin Japan,Mahometans in Turkey, Papistsin Spain,Prot- doctrines to be truebecausecontained in thatbook" (Stephen1962,p.
estantsin England,and Lutherans in Sweden"(p. 638). Yet Locke's 117). Yet, on theotherside of thefence,one mayfindsuchboastsof
editor,Alexander CampbellFraser,writes ina footnote thatwe"cannot circularity as thefollowing: The NewTestament "pretends to containa
Kruskal:Miracles and Statistics 937

divine revelation; and, in support of this high pretension,it gives a we seek for nothingfurther.If only one flower,we seek for nothing
narrativeof the veryfactswhich . . . constitutethe evidence that this further-whatthen if two or three, or more? Each successive one is
pretensionis foundedin truth."That statementis by JohnCook, pro- multipleevidence-proof notadded to proof,butmultipliedbyhundreds
fessorof divinityat St. Mary's College, St. Andrews (Cook 1821, pp. or thousands" (pp. 285-286).
351-352). Variantsof Tindal's criticismcome fromtheinternalstructure 19. There is also a somewhataloofmathematicalviewpointthatsimply
of the Bible. Frye (1982) asks, "How do we knowthatthe Gospel story avoids the verisimilitudequestions emphasized here. Kac (1982), for
is true? Because it confirmsthe propheciesof the Old Testament.But example, writesthat"Statisticalindependence,once a shadowypartner
how do we know thatthe Old Testamentpropheciesare true?Because of gamblers,experimentalscientistsand statisticians,has achieved the
theyare confirmedby the Gospel story.... The two testamentsform respectabilitythat only an ancient disciplinelike number theorycan
a double mirror,each reflecting theother"[fromthereviewbyKermode bestow" (p. 72).
(1982, p. 32)]. 20. Perhapsone reason formisunderstandings about independenceis
A thirdcircularargumentis criticizedby Mill (1969). Mill says,as part inadequate trainingin classrooms and lectures. In a modest probe, I
of hisrecapitulationofHume, that"the existenceof God cannotpossibly oftheAmerican
lookedattwoissuesoftheJournal Association
Statistical
be proved by miracles,forunless a God is alreadyrecognized,the ap- and counted 11 reviewsof introductory textbooks.I inspectedthe six of
parent miracle can always be accounted for on a more probable hy- these books in our libraryand graded theirtreatmentsof independence
pothesis. . .. Once admita God, and theproductionbyhisdirectvolition tolerantly:no A's, one B, two C's, one D, and threeF's, an unhappy
of an effect. .. is no longera purelyarbitrary hypothesis. .. [and] must record. Among the problemswithtreatmentof independencein these
be reckoned with as a serious possibility"(p. 477). In vulgarbrevity, in even findingthe independencediscussions:
books, I cite firstdifficulty
God impliesmiraclesimpliesGod. Indexes were absent or skimpy,and in some cases I gave up aftera
Mill pointsthisout again whenhe writes,"All, therefore, whichHume fruitlesssearch of the index and table of contentstogetherwithrapid
has made out . .. is that. .. no evidencecan prove a miracleto anyone readingin likelychapters.Whentreatments ofindependenceweregiven,
who did not previouslybelieve the existenceof a being or beings with theytendedto be definitionsplus a fewsimplecombinatorialexamples.
supernaturalpower .... If we do not already believe in supernatural Withluck,theremightbe mentionofindependenceforrandomvariables
agencies,no miraclecan prove to us theirexistence"(Mill 1973, vol. 7, and observations.In at most one of the books was thereanythingap-
chap. 25, sec. 2, p. 625). proachingresponsibleconcernforwhat happens when thereis depen-
Stephenmakes a similarpointin discussingHume and Paley: "When dence, how to recognizeit, what analyticoptionsthereare, and so on.
Paley calmlysays,ifwe believe in God, thereis no difficulty in believing 21. An example of extraterrestrial calculationis the correspondence
[in] miracles,Hume's answeris plain. If God is thecause of order,belief between Donaldson and Pollard (1980). Lyttle(1982) provides an ex-
in himdoes not facilitatebeliefin miracles"(Stephen 1962,vol. 1, chap. ample of the blitheassumptionof independenceto calculate the prob-
6, sec. 34, p. 287; see also chap. 8, p. 352). abilityof accidentallaunch of a strategicmissile. For a fine treatment
Lewis, in his well-knownbook on miracles,makes a virtueof the of the dependence problem in estimatingaccident probabilitiesfor li-
circle,withoutmentionof Stephen or Mill, when he writes"Theology quifiednaturalgas transportation see Fairley(1977, especiallypp. 339-
says in effect,'Admit God and withhim the riskof a fewmiracles,and 346). Related problemshave arisenin examinationof acousticevidence
I in returnwillratify yourfaithin uniformity as regardstheoverwhelming about the assassinationof JohnF. Kennedy(National Research Council
majorityof events" (Lewis 1960, p. 106; p. 109 in 1947 ed.). Lewis's 1982, p. 129). Portnoyand Petersen (1984) criticizea studyof biblical
argumentwas not entirelynew. The famous mathematicalphysicist textsforcasual assumptionsof independencein statisticalanalyses. One
G. G. Stokes had givena variant:"Admit the Existenceof a God, of a could easily go on.
personal God, and the possibilityof a miraclefollowsat once" (Stokes 22. An importantexceptionis the positionof De Finettiand his fol-
1891, p. 24). For discussionsee Mascall (1956, pp. 7-8, 180, 185). lowers. [See De Finetti(1982) foran intenselypersonal statementwith
The quotation may remindus not to sneer at what appear obvious key references.]The trulybasic idea forDe Finettiis not independence
circularitiesof argument.Perhaps we are all forcedwilly-nilly into cir- (plus identicaldistribution) butexchangeability, thatis, invarianceof the
cularitiesthatwe dignifyby callingthem"coherences" or "interconsis- relevantjoint distribution underpermutationsof variables.That option
tencies." There is, however,thisdifference,thatsome of us are willing does not affectthe primarythemeof thispaper: to look criticallyat any
to examine our own circularities,and some simplydeny themwithout simplifying assumption-independence,exchangeability, whatever-and
study. to worryabout consequences when the assumptionfails.
Otherdiscussionsof circularity maybe found.I cite as examplesLewis 23. Discussions of such a device for measuringblood pressure are
(1849, pp. 82-93), Chalmers(1857, p. 385), and Walker(1975, p. 162). givenby Wrightand Dore (1970) and Labarthe,Hawkins,and Reming-
The relevanceof thissubtopicis thatit presentsa methodologicaltangle ton (1973, especiallyp. 553). In additionto decreasingor avoidingde-
different fromthose of dependence of testimonyand referenceset, yet pendence among contiguousobservations,devices of thiskindmay also
alike in the frequencywithwhichtheyare ignored. decrease or avoid sources of bias and unwantedvariation:integerpref-
17. The opprobriumstatementseems to have appeared firstin the erences,observerexpectations,and so on.
second (1846) editionof Mill (1952), althoughI do not findit in the 1848
Americanedition(see Mill 1973,vol. 7, p. 538; vol. 8, app. F). Following [ReceivedOctober1987.RevisedMay1988.1
the opprobriumstatement,thereis a brief,pointeddiscussionof over-
simpleprobabilityideas applied to tribunals.For moreon Mill's distaste REFERENCES
forformalism,includingthatof Frenchmathematical-political thought,
Annet, Peter (1744), The Resurrectionof Jesus,London: M. Cooper.
see Schabas (1983). Eddington(1935, pp. 123-125) used theFrenchlegal
Anscombe,F. J. (1978), "StatisticalAnalysis,Special Problemsof: Out-
example as the basis forhis critiqueof casual independence. liers," in InternationalEncyclopediaof Statistics,eds. W. H. Kruskal
18. Meier and Zabell (1980, especiallypp. 501-502) give a fascinating and J. M. Tanur, New York: Free Press, pp. 1039-1944.
storyof 1867 experttestimonyby two greatAmericanmathematicians, Archbishops'Commission (1958), The Church's Ministryof Healing,
Benjamin and Charles S. Peirce (fatherand son), duringa trialover a London: ChurchInformationBoard.
contestedwill. Apparentlythe independenceassumptionwas used cas- Treatise:
Babbage,Charles(1838),TheNinthBridgewater A Fragment
ually in a contextwhereit could hardlyhold. (2nd ed.), London: JohnMurray.(The firsteditionappeared in 1837.)
Casual multiplication in a legal settinghas a literaryendorsementin Barnes,HarryElmer(1965),An Intellectual
and Cultural ofthe
History
WesternWorld(3rd ed.), New York: Dover Publications.(Firstpub-
Poe's "The Mysteryof Marie Roget" (1904). Poe writesthat"If the feet
lished in 1937.)
of Marie being small, those of the corpse were also small, the increase
Bartholomew,DavidJ. (1984),GodofChance, London:SCM.
ofprobabilitythatthebodywasthatofMariewouldnotbe . . . merely andTheology"(withdiscussion),
Statistics
(1988),"iProbability,
arithmetical,
but . . . highly
geometrical,
or accumulative.
Add to all Journal Ser. A., 151,137-178.
Society,
oftheRoyalStatistical
thisshoessuchas she had been knownto wear . . .. Give us, then, Benson,E. F. (1930),As WeWere/AVictorian Peep-Show,London:
flowersin thehatcorresponding to thosewornbythemissing girl,and Longmans,Green.
938 Journalofthe American StatisticalAssociation,December 1988

Berkson,Joseph,Magath,Thomas B., and Hurn,Margaret(1939), "The Fienberg,Stephen E. (1971), "Randomization and Social Affairs:The
Errorof Estimateof the Blood Cell Count as Made Withthe Hemo- 1970 Draft Lottery,"Science, January22, 255-261.
cytometer,"AmericanJournalof Physiology,128, 309-323. Flew, Antony(1967), "Miracles," in Encyclopediaof Philosophy(Vol.
Berry,R. J. (1986), "What to Believe About Miracles," Nature,322, 5), New York: Macmillan and Free Press, pp. 346-353.
July24, 321-322. Frank,JeromeD. (1978), "The Medical PowerofFaith,"Human Nature,
Bertrand,J. (1888), Calcul des Probabilites,Paris: Gauthier-Villars. 1, August,40-47.
Bethe, A. H. (1976), "The Necessityof FissionPower," ScientificAmer- Frank,Philipp G. (1956), "The Varietyof Reasons forthe Acceptance
ican, 234, January,21-31. of ScientificTheories," in The Validationof ScientificTheories,ed.
Bienayme [Jules] (1838), "Probabilite des Jugementset des Temoin- P. G. Frank, Boston: Beacon, pp. 3-18.
ages," in Extraitsdes Proces-Verbauxdes Seances, Paris: Societe Phi- Frye,Northrop (1982),TheGreatCode: TheBibleand Literature, New
lomatique de Paris, pp. 93-96. York: Harcourt,Brace & Jovanovich.[ReviewedbyKermode (1982).]
Bradburn, Norman M. (1969),TheStructure ofPsychological Well-Being,Galton,Francis(1872), "StatisticalInquiriesIntotheEfficacyofPrayer,"
Chicago: Aldine. The Fortnightly Review,New Series 12, August, 125-135. [Reprinted
Brams,StevenJ.(1980),BiblicalGames:A Strategic Analysis ofStories in Means (1876, pp. 85-106).]
in the Old Testament,Cambridge,MA: MIT Press. (1883),Inquiries IntoHumanFaculty andItsDevelopment, Lon-
BritishMedical Association (1956), "Divine Healing and Co-operation don: Macmillan. (Second editionpublishedin 1907 by Dent.)
Between Doctors and Clergy," memorandumto the Archbishops' Garfield,Eugene (1977), "When Citation Analysis StrikesBall Light-
Commissionon Divine Healing, London: Author. ening," in Essays of an InformationScientist(Vol. 2, 1974-1976),
Brown,Colin (1984), Miraclesand theCriticalMind, Grand Rapids, MI: Philadelphia:ISI Press, pp. 479-490. (Firstappeared in CurrentCon-
Eerdmans. tents,May 12, 1976.)
Brush, Stephen G. (1974), "The PrayerTest," AmericanScientist,62, Garraghan,GilbertJ. (1946), A Guide to HistoricalMethod,ed. J. De-
561-563; Correspondence(1975), 63, 6-7. langlez, New York: FordhamUniversityPress.
Burns,R. M. (1981),TheGreatDebateonMiracles/From Joseph Glanvil Gilkey,Langdon (1982), "The CreationistControversy:theInterrelation
to David Hume, Lewisburg,PA: Bucknell UniversityPress. of Inquiryand Belief," Science, Technology,& Human Values, 77,
Chalmers,Thomas (1842), "On the ConsistencyBetween the Efficacy Summer,67-71.
ofPrayer-and theUniformity ofNature" (Discourseson theChristian Grant,RobertM. (1952),Miracleand NaturalLaw in Graeco-Roman
Revelation Viewedin Connection WithModernAstronomy! To Which and Early ChristianThought,Amsterdam:North-Holland.
WeAddedDiscourses oftheConnection
Illustrative Between Theology Haldane, J. B. S. (1928), "Science and Theology as Art Forms," in
and GeneralScience), in Works(Vol. 7), Glasgow: Collins, pp. 234- Possible Worldsand OtherPapers, New York: Harper & Bros., pp.
262. (The catalog of the BritishMuseum gives a range of 1836-1842 237-252. (Reprinted in 1985 in On Being the RightSize and Other
for the Works; it also appears to say that the Discourses was first Essays, ed. J. M. Smith,New York: OxfordUniversitypress.)
publishedin 1818.) Hartman,David (1976), Maimonides:Torahand PhilosophicQuest,Phil-
- (1857),On theMiraculous andInternalEvidences oftheChristian adelphia: JewishPublicationSocietyof America.
Revelation andtheAuthority ofItsRecords,NewYork:Carter.(First Healy, M. J. R. (1979), "Does Medical StatisticsExist?" BulletinIn
publishedin Edinburghin 1814.) AppliedStatistics, 6, 137-182.
Church of England (1986), The Nature of ChristianBelief (Report of Heller, Joseph (1958), "Maimonides' Theory of Miracles," in Between
Churchof England Bishops), London: ChurchHouse. East and West,ed. A. Alexander, London: East & West Library,pp.
Cook, John(1821),An InquiryIntotheBooksof theNew Testament, 112-127.
Edinburgh:Waugh & Innes. Heyde, C. C., and Seneta, E. (1977), I. J. BienaymelStatisticalTheory
Cournot,A.-A. (1838), "Sur les Applicationsdu Calcul des Chances a Anticipated,New York: Springer-Verlag.
la Statistique Judiciare,"Journalde Math6matiques PuresetAppli- Hooper, George (1689), A Fair and MethodicalDiscussion.. ,London:
quees, 3, 257-334. Chiswell.
Cullipp,PlatonJ. (1969), "The EfficacyofPrayer:a Triple-BlindStudy," (1699), "A Calculationof the Credibilityof Human Testimony,"
Medical Times,97, 201-204. Philosophical Transactions
oftheRoyalSociety, 21, 359-365.(There
Daille, John(Jean) (trans. T. Smithand G. Jekyll)(1843), A Treatise has been disagreementabout the authorshipof thispaper. It appeared
on theRightUse of theFathers(2nd ed.), London: Bohn. (Firstpub- in Hooper's Works,publishedby OxfordUniversityPress in 1757 and
lished in 1631.) 1855.)
De Finetti,Bruno (1982), "Probabilityand My Life," in The Makingof House ofBishops(1986), TheNatureof ChristianBelief,London: Church
ed. J. Gani, New York: Springer-Verlag,
Statisticians, pp. 3-20. House.
Deffeyes,KennethS., and MacGregor,Ian D. (1980), "WorldUranium Hume, David (1975), "Of Miracles," in An EnquiryConcerningHuman
Resources," ScientificAmerican,238, January,66-76. Understanding(3rd ed.), ed. L. A. Selby-Bigge,revised by P. H.
Dietl, Paul (1968), "On Miracles," AmericanPhilosophicalQuarterly, Nidditch,Oxford,U.K.: Clarendon Press, sec. 10, p. 109-131. (First
5, 130-134. publishedin 1748; thisis fromthe posthumous1777 ed.)
Donaldson, Thomas, and Pollard, William G. (1980), "Life on Other Joyce,C. R. B., and Weldon, R. M. C. (1965), "The Objective Efficacy
Planets?" AmericanScientist,68, 128-131. ofPrayer:a Double-BlindClinicalTrial,"Journalof ChronicDiseases,
Eddington,Arthur(1935), New Pathwaysin Science, New York: Mac- 18, 367-377.
millan. Kac, Mark (1982), "The Search forthe Meaning of Independence," in
Eggleston, Richard (1978), Evidence, Proof and Probability,London: The Making of Statisticians,
ed. J. Gani, New York: Springer-Verlag,
Weidenfeld& Nicolson. pp. 62-72.
Eichrodt, Walther (trans. J. A. Baker) (1967), Theologyof the Old Kermode, Frank (1982), Review of The Great Code: The Bible and
Testament(Vol. 2), Philadelphia: Westminster.[Published in 1964 Literature,by N. Frye, The New Republic, 186, June9, 30-33.
as Theologie des Alten Testament(5th ed.), Stuttgart:Ehrenfried Kiefer,J. (1979), "Commentson Taxonomy,Independence,and Math-
Klotz.] ematical Models (with referenceto a methodologyof Machol and
Eliade, Mircea (trans. M. L. Ricketts) (1981), Autobiography,Vol. I: Singer)," Mycologia,71, 343-378.
1907-1937,Journey East, Journey West,San Francisco:Harper& Koblitz, Neal (1981), "Mathematics as Propaganda," in Mathematics
Row. Tomorrow,ed. L. A. Steen, New York: Springer-Verlag, pp. 111-
Fairley, William B. (1977), "Evaluating the 'Small' Probabilityof a 120.
CatastrophicAccident From the Marine Transportationof Liquified Kramer,Jane(1985), "LetterFromEurope," TheNew Yorker,61, April
Natural Gas," in Statisticsand Public Policy, by W. Fairley and F. 15, 73-89.
Mosteller,Reading, MA: Addison-Wesley,pp. 331-353. Kruskal, William H. (1960), "Some Remarks on Wild Observations,"
Fairley,WilliamB., and Mosteller,Frederick(1974), "A Conversation Technometrics, 2, 1-3.
AboutCollins,"TheUniversity ofChicagoLaw Review,41,242-253. Labarthe, Darwin R., Hawkins, C. Morton, and Remington,Richard
[Reprinted,along withother relevantmaterials,in 1977 in Statistics D. (1973), "Evaluation of Performanceof Selected Devices forMea-
andPublicPolicy,
Reading,MA: Addison-Wesley, pp. 369-379.] BloodPressure,"
suring TheAmerican Journal
ofCardiology,
32,546-
Falk,Bernard(1942),TheBridgewater CandidFamilyHis-
Millions/A 553.
tory,
London:Hutchinson.[See the(anonymous)reviewofthisbook Lapp,RalphE. (1974),"NuclearSalvationorNuclearFolly?"TheNew
in TheTimesLiterary
Supplement,41,June13,294.] YorkTimesMagazine, February 10, 10-73.
Kruskal:Miracles and Statistics

Lewis, C. S. (1960), Miracles/APreliminary Study(2nd ed.), New York: Perrin, Norman(1977),TheResurrection According to Matthew, Mark,
Macmillan. (Firstpublishedin 1947.) and Luke, Philadelphia:Fortress.
Lewis, George Cornewall(1849), An Essay On theInfluenceofAuthority Poe, Edgar Allan (1904), "The Mysteryof Marie Roget," in Works(Vol.
in Matters of Opinion,London:Parker. - 1), New York: Collier, pp. 248-327. (Firstpublishedin 1842.)
Lewis, Harold W. (1980), "The Safetyof Fission Reactors," Scientific Pomeroy, Ralph S. (1962), "Hume on the Testimonyfor Miracles,"
American,242, March, 53-65. SpeechMonographs, 29, 1-12.
Locke, John(1959), An Essay ConcerningHuman Understanding, ed. Portnoy,Stephen(1988), "Statisticsand ReligiousStudies," in Encyclo-
A. C. Fraser, New York: Dover. (First published in 1690; Fraser pedia of StatisticalSciences (Vol. 8), eds. S. Kotz and N. L. Johnson,
editionpublishedin 1894.) New York: JohnWiley,pp. 679-683.
Lorenz, Konrad Z. (1974), "Analogy as a Source of Knowledge," Sci- Portnoy,StephenL., and Petersen,David L. (1984), "Biblical Textsand
ence, 185, July19, 229-234. StatisticalAnalysis: Zechariah and Beyond," Journalof Biblical Lit-
Lyttle,Bradford(1982), "The Apocalypse Equation," Harvard Maga- erature,103, 11-21.
zine, 89, March-April,19-20. Price, Richard (1777), "On the Importanceof Christianity, the Nature
Machol, RobertE. (1975), "The TitanicCoincidence,"Interfaces, 5, 53- of HistoricalEvidence, and Miracles," in Four Dissertations(4th ed.),
54. London: Cadell, Dissertation4. (Firstpublishedin 1767.)
(1980), "Comments on Taxonomy-A Response," Mycologia, Prior, A. N. (1955), "Can Religion Be Discussed?" in New Essays in
72, 1222-1224. Philosophical Theology, eds. A. Flew and A. Maclntyre,London:
Madansky,Albert (1986), "On Biblical Censuses," Journalof Official SCM, pp. 1-11. [Firstappeared in AustralianJournalof Philosophy
Statistics,2, 561-569. (1942).]
Mahalanobis, P. C. (reportedby J. Tucker, Jr.) (1947), "Summaryof Ramsey, I. T. (1971), "Our Understandingof Prayer," Archbishops'
Lecture on the Combinationof Data From Tests Conducted at Dif- Commissionon ChristianDoctrineOccasionalPaper 1, London: SPCK.
ferentLaboratories,"ASTM Bulletin,No. 144, 63-65. Romanes,Ethel(1908),TheLifeand Letters ofGeorgeJohnRomanes
Maimonides, Moses (trans. S. Pines) (1963), The Guide of the Per- (6th impression),London: Longmans,Green.
plexed,Chicago:Universityof Chicago Press. Romanes, George J. (1874), ChristianPrayerand General Laws, Lon-
Marshall,Eliot (1986), "Academy MembershipFightGoes Public," Sci- don: Macmillan. (Includes the BurneyPrize essay for 1873.)
ence, 234, 1192-1194; Follow-up (1987), 236, 661-662. Rousseau, JeanJacques(trans.B. Foxley) (1911), Emile, London: Deut.
Marty,MartinE. (1977), "Science Versus Religion: An Old Squabble [Firstpublishedin 1762. The quotation is from"The Creed of a Sa-
SimmersDown," SaturdayReview,5, December 10, 29-35. voyardPriest" (book 4).]
Mascall, E. L. (1956), ChristianTheologyand Natural Science, New Schabas, Margaret(1983), "JohnStuartMill to WilliamStanleyJevons:
York: Ronald. An UnpublishedLetter," The Mill News Letter,18, No. 2, 24-28.
Mason, Brian (1967), "Meteorites," AmericanScientist,55, 429-455. Smart,Ninian (1964), Philosophersand ReligiousTruth,London: SCM.
McArthur,Harvey K. (1965), "A Surveyof Recent Gospel Research," Sobel, JordanHoward (1987), "On the Evidence of TestimonyforMir-
in New TheologyNo. 2, eds. M. E. Martyand D. G. Peerman,New acles: A Bayesian Interpretationof David Hume's Analysis," The
York: Macmillan,pp. 201-221. (Firstpublishedin 1964.) Philosophical Quarterly, 37, 166-186.
McCormick,Charles Tilford(1972), Handbook of theLaw of Evidence Spence, JanetT., and Helmreich,RobertL. (1978), Masculinity& Fem-
(2nd ed.), ed. E. W. Cleary, St. Paul, MN: West Publishing. ininitylTheirPsychological Dimensions, Correlates, & Antecedents,
McKinnon, Alastair (1967), "'Miracle' and 'Paradox'," AmericanPhil- Austin: Universityof Texas Press.
osophical Quarterly,4, 308-314. Stead, W. T. (1891), "The Princeof Wales," The Reviewof Reviews,4,
Means, John0. (ed.) (1876), The Prayer-GaugeDebate, Boston: Con- 23-34. [Ascriptionof authorshipcomes fromBenson (1930, p. 211)
gregationalPublishingSociety. and frominternalevidence.]
Meier, Paul, and Zabell, Sandy (1980), "Benjamin Peirce and the How- Stephen,Leslie (1962), Historyof EnglishThought in theEighteenth
land Will," Journalof theAmericanStatisticalAssociation,75, 497- Century(3rd ed.), New York: Harcourt,Brace & World. (Firstpub-
506. lished in 1876.)
Mill, John Stuart (1885), Nature, the Utilityof Religion and Theism, Stevens, William Arnold, and Burton, Ernest DeWitt (1905), A Har-
London: Longmans,Green. [Firstpublishedin 1874. For a fullversion monyoftheGospelsforHistorical Study(4thed.), NewYork:Charles
see the 1969 Toronto edition, subtitled"Three Essays on Religion" Scribner.
(Vol. 10 of the CollectedWorks).] Stewart,Balfour (1878), "Science and a Future State," The Princeton
(1952), A Systemof Logic/Ratiocinative and Inductive(8th ed.), Review,54, 399-408.
London: Longmans,Green. [Firstpublishedin 1843. Second edition Stigler,Stephen(1986), "JohnCraigand theProbabilityofHistory:From
1846. For fullscholarlytreatmentsee the 1973 edition (Vol. 7), Uni- the Death of Christto the Birthof Laplace," Journalof theAmerican
versityof Toronto Press.] Statistical
Association, 81, 879-887.
(1969), CollectedWorks,Toronto: Universityof Toronto Press. Stokes, G. G. (1891), Natural Theology,London: Black.
(1973), A Systemof Logic/Ratiocinative and Inductive,Toronto: Strauss,David Friedrich[trans.G. Eliot (M. Evans)] (1892), The Life
Universityof Toronto Press. of Jesus, London: Swan Sonnenschein. (First published in 1835, in
National Research Council (1982), "Reexamination of Acoustic Evi- Tuibingen,as Das Leben Jesu.)
dence in the Kennedy Assassination,"Science, 218, October 8, 127- Swerdlow,Noel M. (1979), "Ptolemyon Trial," The AmericanScholar,
133; Correction(1982), 218, November 15, 521. 48, 523-531. (Essay-reviewof R. R. Newton's The Crimeof Claudius
Owen, David (1987), "Hume Versus Price on Miracles and PriorProb- Ptolemy,JohnsHopkins UniversityPress.)
abilities:Testimonyand the Bayesian Calculation," The Philosophical Swinburne,R. G. (1968), "Miracles," PhilosophicalQuarterly,18, 320-
Quarterly,37, 187-202. 328.
Page, E. S. (1978), "Process Control," in InternationalEncyclopediaof (1970), The Conceptof Miracle,London: Macmillan.
Statistics,eds. W. H. Kruskaland J. M. Tanur,New York: Free Press, Thomas, Jo (1985), "Toward the BluntBishop, ThatcherIs Irreverent,"
pp. 809-812. NewYorkTimes,May9.
Patey,Douglas Lane (1984), Probabilityand LiteraryForm/Philosophic Thompson,Harold William(1931), A ScottishMan of Feeling,London:
Theoryand LiteraryPracticein theAugustanAge, Cambridge,U.K.: OxfordUniversityPress.
CambridgeUniversityPress. Trench, Richard Chenevix (1860), Notes on the Miracles of Our Lord
Pearson, Karl (1924), "Historical Note on the Origin of the Normal (2nd Amer. ed.), New York: Appleton.
Curve of Error," Biometrika,16, 402-404. Trinklein,FrederickE. (1971), The God of Science,Grand Rapids, MI:
(1924-1930), The Life, Lettersand Labours of Francis Galton, Eerdmans.
Cambridge,U.K.: CambridgeUniversityPress. Turner,Frank M. (1974), "Rainfall, Plagues, and the Princeof Wales:
Pearson, Karl (1978), The Historyof Statisticsin the17thand 18thCen- A Chapterin the Conflictof Religionand Science," Journalof British
turies,ed. E. S. Pearson, London: CharlesW. Griffin.(Lecturesgiven Studies,13,46-65.
in 1921-1933.) Tyndall,John(1871), "Prayerand NaturalLaw," in Fragments ofScience
Peirce,C. 5. (1958),"4The
LawsofNatureandHume'sArgument People,
Against forUnscientific London:Longmans, Green,pp. 31-38. [First
Miracles,"Valuesina UniverseofChance/Selected ofCharles
Writings in Mountaineering
published (1861).]
S. Peirce(1839-1914),ed. P. P. Wiener,Stanford, Uni- Uman,M. A. (1982),ReviewofBallLightening
CA: Stanford andBeadLightening,
versity Press,pp. 289-321.(Writtenin 1901.) byJ.D. Barry,American 80, 535-536.
Scientist,
940 Journalofthe American StatisticalAssociation,December 1988

Van Leeuwen,HenryG. (1963), TheProblemof Certainty in English Wright,B. M., and Dore, C. F. (1970),"A Random-Zero Sphygmo-
ThoughtIl630-1690, theHague: Nijhoff. manometer," TheLancet,1, February14,337-338.
Venn,John(1962),TheLogicof Chance(4thed.), NewYork:Chelsea Young,J. R. (1844),An Elementary Treatiseon Algebra,Theoretical
Publishing. in 1866;3rded. 1888.)
(Firstpublished and Practical(4thed.), London:Souter& Law.
Walker, D. P. (1975),Spiritual
andDemonicMagic/From FicinotoCam- (1846),ThreeLecturesAddressedtotheStudents College
ofBelfast
panella,SouthBend,IN: University ofNotreDame Press.(Firstpub- on SomeoftheAdvantages ofMathematicalStudyto WhichIs Added
lishedin 1958.) an Examination ofHume'sArgument London:Sou-
AgainstMiracles,
Westrum, Ron (1978),"Scienceand SocialIntelligence AboutAnom- ter & Law.
alies: The Case of Meteorites," Analysisof Testimony,"
Social Studiesof Science,8, 461- Zabell, S. L. (in press),"The Probabilistic
493. Journal Planning
ofStatistical andInference,20.

Das könnte Ihnen auch gefallen