Barnard 1958

Biometrika Trust
Studies in the History of Probability and Statistics: IX. Thomas Bayes's Essay Towards Solving
a Problem in the Doctrine of Chances
Author(s): G. A. Barnard and Thomas Bayes
Reviewed work(s):
Source: Biometrika, Vol. 45, No. 3/4 (Dec., 1958), pp. 293-315
Published by: Biometrika Trust
Stable URL: http://www.jstor.org/stable/2333180 .
Accessed: 02/01/2013 16:27
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
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.
Biometrika Trust is collaborating with JSTOR to digitize, preserve and extend access to Biometrika.
http://www.jstor.org
This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

All use subject to JSTOR Terms and Conditions
VOLUME 45, PARTS 3 AND 4 DECEMBER 1958
STUDIES IN THE HISTORY OF PROBABILITY AND STATISTICS

IX. THOMAS BAYES'S ESSAY TOWARDS SOLVING A PROBLEM
IN THE DOCTRINE OF CHANCES*
[Reproduced with the permissionof the Council of the Royal Society from
The PhilosophicalTransactions(1763), 53, 370-418]
THOMAS BAYES-A BIOGRAPHICAL NOTE
BY G. A. BARNARD
ImperialCollege,London
Bayes's paper,reproducedin the following pages,mustrankas one of the mostfamous

memoirsin the historyofscienceand the problemit discussesis stillthe subjectofkeen
controversy. statureofBayeshimself
The intellectual is measuredbythefactthatit is still
interestto knowwhatBayeshad to say onthequestionshe
as wellas historical
ofscientific
raised. And yetsuchare the vagariesofhistoricalrecords,thatalmostnAothing is known
about thepersonalhistoryoftheman.TheDictionary ofNationalBiography, compiledat
theendofthelastcentury, whenthewholetheoryofprobability wasintemporary eclipsein
England, has an entrydevoted to Bayes's father,Joshua Bayes, F.R.S., one of the firstsix
Nonconformistministersto be publicly ordained as such in England, but it has nothingon
his much more distinguishedson. Indeed, the note on Thomas Bayes whichis to appear in
the forthcomingnew edition of the Encyclopedia Britannica will apparently be the first
biographical note on Bayes to appear in a work of general referencesince the Imperial
Dictionaryof UniversalBiographywas publishedin Glasgow in 1865. And in treatiseson the
historyofmathematics,such as that ofLoria (1933) and Cantor (1908), noticeis taken ofhis
contributionsto probabilitytheoryand to mathematicalanalysis, but biographicaldetails
are lacking.
The Reverend Thomas Bayes, F.R.S., author of the firstexpressionin precise,quantita-
tive formof one of the modes of inductive inference,was born in 1702, the eldest son of
Ann Bayes and Joshua Bayes, F.R.S. He was educated privately,as was usual with Non-
conformistsat that time, and fromthe fact that when Thomas was 12 Bernoulli wrote to
Leibniz that 'poor de Moivre' was having to earn a livingin London by teaching mathe-
matics,we are temptedto speculate that Bayes may have learned mathematicsfromone of
the foundersofthe theoryofprobability. Eventually Thomas was ordained,and began his
ministryby helpinghis father,who was at the time stated, ministerof the Presbyterian
meetinghouse in Leather Lane, offHolborn. Later the son went to ministerin Tunbridge
Wells at the PresbyterianChapel on Little Mount Sion whichhad been opened on 1 August
1720. It is not known when Bayes went to Tunbridge Wells, but he was not the firstto
ministeron Little Mount Sion,and he was certainlytherein 1731,whenhe produced a tract
entitled 'Divine Benevolence, or an attemptto prove that the PrincipleEnd of the Divine
* Thomas Bayes's famousEssay is so oftenreferredto in currentstatisticalliterature,but so rarely
studied because of the difficulty
of access, that the Editors have feltjustifiedin reprintingit in the
BiometrikaHistoryof Probabilityand Statisticsseries.
I9 Biom. 45

294 Studie8in thehi8tory
ofprobability
and 8tatistiC8.
IX
Providence and Governmentis the happiness ofHis Creatures'. The tractwas publishedby
John Noon and copies are in Dr Williams's libraryand the BritishMuseum. The following
is a quotation:
[p. 22]: I don't find (I am sorryto say it) any necessaryconnectionbetween mere intelligence,
thoughever so great, and the love or approbation of kind and beneficentactions.
Bayes argued that the principal end of the Deity was the happiness of His creatures,in
opposition to Balguy and Grove who had, respectively,maintainedthat the firstspringof
action of the Deity was Rectitude, and Wisdom.
In 1736JohnNoon publisheda tractentitled'An Introductionto theDoctrineofFluxions,
and a Defence of the Mathematiciansagainst the objections of the Authorof the Analyst'.
De Morgan (1860) says: 'This very acute tract is anonymous,but it was always attributed
to Bayes by the contemporarieswho writein the names of the authors as I have seen in
various copies, and it bears his name in otherplaces.' The ascriptionto Bayes is accepted
also in the BritishMuseum catalogue.
From the copy in Dr Williams's librarywe quote:
[p. 9]: It is not the businessoftheMathematicianto disputewhetherquantitiesdo in factevervaryin
the mannerthat is supposed,but onlywhetherthe notionof theirdoingso be intelligible;whichbeing
allowed,he has a rightto take it forgranted,and thensee what deductionshe can make fromthat sup-
position. It is not the businessofa Mathematicianto show that a straitline or circlecan be drawn,but
he tellsyou whathe meansby these;and ifyou understandhim,you may proceedfurther withhim; and
it would not be to the purpose to object that thereis no such thingin nature as a true straitline or
perfectcircle,forthis is none of his concern:he is not inquiringhow thingsare in matterof fact,but
supposingthingsto be in a certainway,whatare the consequencesto be deducedfromthem;and all that
is to be demanded of him is, that his suppositionsbe intelligible,and his inferencesjust fromthe sup-
positionshe makes.
[p. 48]: He [i.e. the Analyst = Bishop Berkeley] representsthe disputes and controversiesamong
mathematiciansas disparagingthe evidenceoftheirmethods:and, Query51, he representsLogics and
Metaphysicsas properto open theireyes,and extricatethemfromtheirdifficulties. Now wereevertwo
thingsthus put together?If the disputes of the professorsof any science disparage the science itself,
Logics and Metaphysicsare muchmoreto be disparagedthan Mathematics;why,therefoTe, ifI am half
blind,mustI take formy guide one that can't see at all?
[p. 50]: So faras Mathematicsdo not tend to make men moresoberand rationalthinkers,wiserand
bettermen,theyare onlyto be consideredas an amusement,whichoughtnot to take us offfromserious
business.
This tractmay have had somethingto do withBayes's election,in 1742,to Fellowship ofthe
Royal Society, forwhich his sponsorswere Earl Stanhope, MartinFolkes, James Burrow,
CromwellMortimer,and John Eames.
William Whiston, Newton's successor in the Lucasian Chair at Cambridge, who was
expelled fromthe UniversityforArianism,notesin his Memoirs(p. 390) that 'on Augustthe
24ththisyear 1746,beingLord's Day, and St. Bartholomew'sDay, I breakfastedatMr Bay's,
a dissenting Ministerat Tunbridge Wells, and a Successor, though not immediate, to
Mr HumphreyDitton, and like him a verygood mathematicianalso'. Whiston goes on to
relate what he said to Bayes, but he gives no indicationthat Bayes made reply.
Accordingto Strange (1949) Bayes wished to retirefromhis ministryas early as 1749,
whenhe allowed a groupofIndependentsto bringministersfromLondon to take servicesin
his chapel week by week,except forEaster, 1750,when he refusedhis pulpit to one of these
preachers;and in 1752 he was succeededin his ministryby theRev. WilliamJohnston,A.M.,
who inheritedBayes's valuable library. Bayes continued to live in TunbridgeWells until
his death on 17 April 1761. His body was taken to be buried,withthat ofhisfather,mother,

G. A. BARNARD 295
brothersand sisters,in the Bayes and Cotton familyvault in Bunhill Fields, the Noncon-
formistburialgroundby Moorgate. This cemeteryalso containsthe grave of Bayes's friend,
the Unitarian Rev. Richard Price, author of the NorthamptonLife Table and object of
Burke's oratory and invective in Reflectionson the French Revolution,and the graves
of John Bunyan, Samuel Watts, Daniel Defoe, and many otherfamous men.
Bayes's will,executed on 12 December 1760,showshimto have been a man of substance.
The bulk of his estate was divided among his brothers,sisters,nephewsand cousins,but he
left?200 equally between 'John Boyl late preacherat Newingtonand now at Norwich,and
Richard Price now I suppose preacherat NewingtonGreen'. He also left'To Sarah Jeffery
daughter of John Jeffery,living with her father at the corner of Fountains Lane near
TonbridgeWells, ?500, and mywatch made by Elliott and all mylinenand wearingapparell
and household stuff.'
Apart fromthe tracts already noted, and the celebrated Essay reproducedhere, Bayes
wrotea letteron AsymptoticSeriesto JohnCanton,publishedin the PhilosophicalTransac-
tions of the Royal Society (1763, pp. 269-271). His mathematical work, though small in
quantity,is of the veryhighestquality; both his tract on fluxionsand his paper on asymp-
totic series contain thoughtswhich did not receive as clear expressionagain until almost
a centuryhad elapsed.
Since copies ofthevolumein whichBayes's essay firstappeared are not rare,and copies of
a photographicreprintissued by the DepartmentofAgriculture,Washington,D.C., U.S.A.,
are fairlywidely dispersed,the view has been taken that in preparingBayes's paper for
publicationheresomeeditingis permissible.In particular,thenotationhas been modernized,
some ofthe archaismshave been removedand what seemto be obvious printer'serrorshave
been corrected.Sometimes,whena word has been omittedin the original,a suggestionhas
been supplied,enclosedin square brackets. Otherwise,however,nothinghas been changed,
and we hope that while the present text should in no sense be regarded as definitive,it
will be easier to read on that account. All the work ofpreparingthe text forthe printerwas
most painstakinglyand expertlycarriedout by Mr M. Gilbert,B.Sc., A.R.C.S. Thanks are
also due to the Royal Society forpermissionto reproduce the Essay in its presentform.
In writingthe biographicalnotes the presentauthor has had the friendlyhelp of many
persons, includingespecially Dr A. Fletcher and Mr R. L. Plackett, of the Universityof
Liverpool, Mr J. F. C. Willder, of the Department of Pathology, Guy's Hospital Medical
School, and Mr M. E. Ogborn,F.I.A., of the Equitable Life Assurance Society. He would
also like to thank Sir Ronald Fisher, forsome initial proddingwhich set him moving,and
Prof. E. S. Pearson, forpatient encouragementto see the matterthroughto completion.
REFERENCES
ANDERSON J. G. (1941). MathematicalGazette,25, 160-2.
CANTOR,M. (1908). Geschichte derMathematik,vol. iv. (Article by Netto.)
DE MORGAN,A. (1860). Notes and Queries,7 Jan. 1860.
LORIA, G. (1933). Storia delle Matematiche,vol. III. Turin.
MACKENZIE, M. (Ed.) (1865). Imperial Dictionaryof UniversalBiography,3 vols. Glasgow.
STRANGE, C. H. (1949). Nonconformity in TunbridgeWells. TunbridgeWells.
The Gentleman'sMagazine (1761). 31, 188.
Notes and Queries (1941). 19 April.

[ 296 ]
AN ESSAY TOWARDS SOLVING A PROBLEM IN THE

DOCTRINE OF CHANCES
BY THE LATE REV. MR BAYES, F.R.S.

Communicatedby Mr Price, in a Letterto John Canton,A.M., P.R.S.
Read 23 December1763
Dear Sir,
I now send you an essay which I have found among the papers of our deceased friend
Mr Bayes, and which,in my opinion, has great merit,and well deserves to be preserved.
Experimentalphilosophy,you will find,is nearlyinterestedin the subject ofit; and on this
account thereseems to be particularreason forthinkingthat a communicationof it to the
Royal Society cannot be improper.
He had, you know,the honourofbeinga memberofthat illustriousSociety,and was much
esteemed by many in it as a very able nmathematician.In an introductionwhich he has
writto this Essay, he says, that his design at firstin thinkingon the subject of it was, to
findout a methodby whichwe mightjudge concerningthe probabilitythat an event has to
happen, in given circumstances,upon suppositionthat we know nothingconcerningit but
that, under the same circumstances,it has happened a certainnumberof times,and failed
a certainothernumberof times. He adds, that he soon perceivedthat it would not be very
difficultto do this, provided some rule could be found according to which we ought to
estimate the chance that the probabilityforthe happeningof an event perfectlyunknown,
should lie betweenany two named degreesofprobability,antecedentlyto any experiments
made about it; and that it appeared to him that the rule must be to suppose the chance the
same that it should lie betweenany two equidifferent degrees; which,ifit were allowed, all
the rest mightbe easily calculated in the common method of proceedingin the doctrineof
chances. Accordingly,I findamong his papers a veryingenioussolution of this problemin
thisway. But he afterwardsconsidered,that thepostulateon whichhe had argued mightnot
perhaps be looked upon by all as reasonable; and thereforehe chose to lay down in another
formthe propositionin which he thoughtthe solution of the problemis contained,and in
as choliumto subjoin the reasons why he tho-ughtso, ratherthan to take into his mathe-
matical reasoningany thingthat mightadmit dispute. This, you will observe,is the method
which he has pursued in this essay.
Every judicious personwill be sensiblethat the problemnow mentionedis by no means
merelya curious speculation in the doctrineof chances, but necessaryto be solved in order
to [provide]a surefoundationforall our reasoningsconcerningpast facts,and what is likely
to be hereafter.Commonsense is indeed sufficient to shew us that, fromthe observationof
what has in formerinstancesbeen the consequenceofa certaincause or action,one may make
a judgmentwhat is likelyto be the consequence of it anothertime,and that the larger[the]
numberof experimentswe have to supporta conclusion,so much the morereason we have
to take it forgranted. But it is certainthat we cannot determine,at least not to any nicety,
in what degreerepeatedexperimentsconfirma conclusion,withoutthe particulardiscussion
of the beforementionedproblem; which, therefore,is necessary to be considered by any

THOMAS BAYES 297
of analogical or inductive
one who wouldgive a clearaccountof the strength reasoning;
we seemto knowlittlemorethanthatit doessometimes
whichat present,
concerning,
us,andat othertimesnot;andthat,as it is themeansof[a]cquainting
infactconvince
uswithmanytruths,ofwhichotherwisewemusthavebeenignorant; so itis,inallproba-
bility,the source of many errors,which perhaps mightin some meas:re be avoided, if the
force that this sort of reasoning ought to have with us were more distinctlyand clearly
understood.
These observationsprovethat theproblemenquiredafterin thisessay is no less important
than it is curious. It may be safelyadded, I fancy,that it is also a problemthat has never
beforebeen solved. Mr De Moivre,indeed, the great improverof this part of mathematics,
has in his Laws ofChance,* afterBernoulli,and to a greaterdegreeofexactness,given rules
to findthe probabilitythereis, that ifa verygreatnumberoftrialsbe made concerningany
event,the proportionofthe numberoftimesit willhappen,to the numberoftimesit willfail
in those trials, should differless than by small assigned limitsfromthe proportionof the
probabilityofits happeningto the probabilityofits failingin one singletrial. But I know of
no person who has shewn how to deduce the solution of the converse problem to this;
namely, 'the numberof times an unknownevent has happened and failed being given, to
findthe chance that the probabilityofits happeningshould lie somewherebetweenany two
named degreesof probability.' What Mr De Moivre has done thereforecannot be thought
sufficient to make the considerationofthispointunnecessary:especially,as the ruleshe has
given are not pretendedto be rigorouslyexact, except on suppositionthat the numberof
trialsmade are infinite;fromwhenceit is not obvious how largethe numberoftrialsmustbe
in orderto make them exact enough to be depended on in practice.
MrDe Moivrecalls theproblemhe has thus solved,the hardestthat can be proposedon the
subject of chance. His solution he has applied to a very importantpurpose,-and thereby
shewn that those are much mistakenwho have insinuatedthat the Doctrine of Chances in
mathematicsis oftrivialconsequence,and cannot have a place in any seriousenquiry.t The
purposeI mean is, to shewwhat reason we have forbelievingthat thereare in the constitu-
tion of thingsfixtlaws accordingto whicheventshappen, and that,therefore,the frameof
the world must be the effectof the wisdom and power of an intelligentcause; and thus to
confirmthe argumenttaken fromfinalcauses forthe existenceofthe Deity. It will be easy
to see that the converse problem solved in this essay is more directlyapplicable to this
purpose; forit shewsus, withdistinctnessand precision,in everycase ofany particularorder
or recurrencyofevents,what reasonthereis to thinkthat such recurrencyor orderis derived
from stable causes or regulations in natureX, and not from any of the irregularitiesof
chance.
The two last rulesin thisessay are givenwithoutthe deductionsofthem. I have chosento
do this because these deductions,taking up a good deal ofroom,would swell the essay too
much; and also because theserules,thoughofconsiderableuse, do not answerthepurposefor
whichtheyare givenas perfectlyas could be wished.They are howeverreadyto be produced,
if a communicationof them should be thoughtproper. I have in some places writ short
notes, and to the whole I have added an application of the rules in the essay to some
* See MrDe Moivre'sDoctrineofChances,p. 243,etc. He has omittedthe demonstrations ofhis rules,
but these have been since supplied by Mr Simpsonat the conclusionof his treatiseon The Natureand
Laws ofChance.
t See his Doctrineof Chances,p. 252, etc.

298 Studiesin thehistoryofprobability
and statistics.IX
particularcases, in orderto conveya cleareridea of the natureof the problem,and to shew
how farthe solutionof it has been carried.
I am sensiblethat yourtimeis so muchtaken up that I cannotreasonablyexpect that you
should minutelyexamine every part of what I now send you. Some of the calculations,
particularlyin the Appendix,no one can make withouta good deal oflabour. I have taken
so much care about them,that I believe therecan be no materialerrorin any of them; but
should therebe any such errors,I am the onlypersonwho oughtto be consideredas answer-
able forthem.
Mr Bayes has thoughtfitto begin his workwitha briefdemonstrationofthe generallaws
of chance. His reason fordoing this,as he says in his introduction,was not merelythat his
reader mightnot have the troubleof searchingelsewhereforthe principleson whichhe has
argued,but because he did not knowwhitherto referhimfora clear demonstrationof them.
He has also made an apology forthe peculiar definitionhe has given of the word chanceor
probability.His designhereinwas to cut offall disputeabout the meaningofthe word,which
in commonlanguageis used in different sensesby personsofdifferent opinions,and according
as it is applied to past or futurefacts. But whatever differentsenses it may have, all (he
observes) wi]l allow that an expectation depending on the truth of any past fact, or the
happeningofanyfutureevent,oughtto be estimatedso much the morevaluable as-thefact
is morelikelyto be true,or the event morelikelyto happen. Instead therefore, ofthe proper
sense ofthe wordprobability,he has given that whichall will allow to be its propermeasure
in every case wherethe word is used. But it is time to conclude this letter. Experimental
philosophy is indebted to you for several discoveries and improvements;and, therefore,
I cannot help thinkingthat there is a peculiar proprietyin directingto you the foliowing
essay and appendix. That yourenquiriesmay be rewardedwithmanyfurthersuccesses,and
that you may enjoy every valuable blessing,is the sincerewish of,Sir,
your very humble servant,
Newington-Green, Richard Price
10 November1763
PROBLEM
Giventhe numberof timesin whichan unknownevent has happened and failed: Required
the chance that the probabilityof its happening in a single trial lies somewherebetween
any two degreesof probabilitythat can be named.
SECTION I
DEFINITION 1. Several events are inconsistent, when if one of them happens, none of the
rest can.
2. Two events are contrarywhen one, or otherof them must; and both togethercaninot
happen.
3. An eventis said tofail, whenit cannothappen; or,whichcomesto the same thing,when
its contraryhas happened.
4. An event is said to be determinedwhen it has eitherhappened or failed.
5. The probabilityof any eventis the ratio between the value at which an expectation
dependingon the happeningof the event ought to be computed,and the value of the thing
expected upon it's happening.

THOMAS BAYES 299
6. By chanceI mean the same as probability.
7. Events are independentwhen the happeningof any one of them does neitherincrease
nor abate the probabilityof the rest.
Prop. 1
When several events are inconsistentthe probabilityof the happeningof one or otherof
them is the sum of the probabilitiesof each of them.
Suppose therebe threesuch events,and whicheverofthemhappens I am to receiveN, and
that the probabilityof the 1st, 2nd, and 3rd are respectivelya/N, b/N,c/N. Then (by the
definitionofprobability)the value ofmy expectationfromthe 1stwill be a, fromthe 2nd b,
and fromthe 3rd c. Whereforethe value of my expectationsfromall threewill be a + b + c.
But the sum of my expectationsfromall threeis in this case an expectation of receivingN
upon the happeningof one or otherof them. Wherefore(by definition5) the probabilityof
one or otherofthemis (a + b + c)/N or a/N + b/N+ c/N. The sum ofthe probabilitiesofeach
of them.
COROLLARY. If it be certain that one or other of the three events must happen, then
a + b + c = N. For in this case all the expectationstogetheramountingto a certainexpecta-
tion ofreceivingN, theirvalues togethermustbe equal to N. And fromhenceit is plain that
the probabilityof an event added to the probabilityof its failure(or of its contrary)is the
ratio of equality. For these are two inconsistentevents, one of whichnecessarilyhappens.
Whereforeif the probabilityof an event is P/N that of it's failurewill be (N - P)/N.
Prop. 2
If a personhas an expectationdependingon the happeningofan event,the probabilityof
the event is to the probabilityof its failureas his loss if it failsto his gain if it happens.
Suppose a person has an expectation of receivingN, dependingon an event the proba-
bilityofwhichis P/N. Then (by definition5) the value ofhis expectationis P, and therefore
if the event fail,he loses that whichin value is P; and if it happens he receives N, but his
expectation ceases. His gain thereforeis N - P. Likewise since the probabilityof the event
is P/N, that ofits failure(by corollaryprop. 1) is (N - P)/N. But P/N is to (N - P)/N as P is
to N - P, i.e. the probabilityof the event is to the probabilityof it's failure,as his loss ifit
fails to his gain ifit happens.
Prop. 3
The probabilitythat two subsequent eventswillboth happen is a ratio compoundedofthe
probabilityof the 1st,and the probabilityof the 2nd on suppositionthe 1st happens.
Suppose that, if both events happen, I am to receive N, that the probabilityboth will
happen is P/N, that the 1st will is a/N (and consequentlythat the 1st will not is (N - a)/N)
and that the 2nd will happen upon suppositionthe 1st does is b/N. Then (by definition5)
P willbe the value ofmyexpectation,whichwillbecome b ifthe 1sthappens. Consequently
if the 1sthappens, mygain by it is b - P, and if it failsmy loss is P. Wherefore,by the fore-
going proposition,a/N is to (N - a)/N, i.e. a is to N - a as P is to b - P. Wherefore(com-
ponendoinverse)a is to N as P is to b. But the ratio of P to N is compoundedof the ratio of P
to b, and that of b to N. Whereforethe same ratio of P to N is compounded of the ratio of
a to N and that ofb to N, i.e. theprobabilitythat thetwo subsequenteventswillboth happen
is compoundedofthe probabilityofthe 1stand theprobabilityofthe 2nd on suppositionthe
1st happens.

300 Studies in thehistoryof probabilityand statistics. IX
COROLLARY. Hence ifoftwo subsequenteventsthe probabilityofthe 1stbe a/N, andthe
probabilityof both togetherbe P/N, then the probabilityof the 2nd on supposition the
1st happens is P/a.
Prop. 4
If therebe twosubsequenteventsto be determinedeveryday,and each day theprobability
ofthe 2nd is b/Nand the probabilityof both P/N, and I am to receiveN ifboth the events
happen the firstday on whichthe 2nd does; I say, accordingto these conditions,the proba-
bilityofmyobtainingN is P/b. For ifnot,let the probabilityof myobtainingN be x/Nand
let ybe to x as N - b to N. Then sincex/NistheprobabilityofmyobtainingN (bydefinition1)
x is thevalue ofmyexpectation.And again,because accordingto theforegoingconditionsthe
firstday I have an expectationofobtainingN dependingon the happeningofboththeevents
together,theprobabilityofwhichis P/N, the value ofthisexpectationis P. Likewise,ifthis
coincident should not happen I have an expectation of being reinstated in my former
circumstances,i.e. of receivingthat whichin value is x dependingon the failureof the 2nd
event the probabilityofwhich (by cor. prop. 1) is (N - b)/Nor y/x,because y is to x as N - b
to N. Whereforesincex is thethingexpectedand y/xtheprobabilityofobtainingit,thevalue
ofthisexpectationis y. But thesetwo last expectationstogetherare evidentlythe same with
my originalexpectation,the value of which is x, and thereforeP + y = x. But y is to x as
N - b is to N. Whereforex is to P as N is to b, and x/N (the probabilityof my obtainingN)
is P/b.
COR. Suppose afterthe expectationgiven me in the foregoing proposition,and beforeit is
at all knownwhetherthe 1steventhas happened or not,I should findthat the 2nd event has
happened; fromhenceI can onlyinferthat the eventis determinedon whichmyexpectation
depended, and have no reason to esteem the value of my expectation eithergreateror less
than it was before.For if I have reason to thinkit less,it would be reasonable forme to give
somethingto be reinstatedin myformercircumstances,and thisoverand overagain as often
as I should be informedthat the 2nd event had happened, whichis evidentlyabsurd. And
the like absurdityplainlyfollowsifyou say I oughtto set a greatervalue on my expectation
than before,forthen it would be reasonable forme to refusesomethingif offeredme upon
condition I would relinquishit, and be reinstatedin my formercircumstances;and this
likewiseover and over again as oftenas (nothingbeing knownconcerningthe 1st event) it
should appear that the 2nd had happened. Notwithstandingthereforethis discoverythat
the 2nd event has happened, my expectation ought to be esteemed the same in value as
before,i.e. x, and consequentlythe probabilityof my obtainingN is (by definition5) still
xIN or P/b.* But afterthis discoverythe probabilityof my obtainingN is the probability
that the 1st oftwo subsequent events has happened upon the suppositionthat the 2nd has,
whoseprobabilitieswereas beforespecified.But theprobabilitythat an eventhas happened
is the same as the probabilityI have to guess rightifI guess it has happened. Whereforethe
followingpropositionis evident.
* What is heresaid may perhapsbe a littleillustratedby consideringthat all that can be lost by the
happeningof the 2nd event is the chance I should have had of being reinstatedin my formercircum-
stances,iftheeventon whichmyexpectationdependedhad been determinedin themannerexpressedin
the proposition.But this chance is always as muchagainst me as it isfor me. If the 1st eventhappens,
it is against me, and equal to the chance forthe 2nd event's failing.If the 1st event does not happen,
it isfor me, and equal also to the chance forthe 2nd event's failing.The loss of it, therefore,
can be no
disadvantage.

THOMAS BAYES 301
Prop. 5
If therebe two subsequent events,the probabilityof the 2nd b/Nand the probabilityof
bothtogetherP/N, and it beingfirstdiscoveredthatthe 2nd event has happened, fromhence
I guess that the 1st event has also happened, the probabilityI am in the rightis P/b.*
Prop. 6
The probabilitythat severalindependenteventsshall all happen is a ratio compoundedof
the probabilitiesof each.
For fromthe nature of independentevents,the probabilitythat any one happens is not
alteredby the happeningor failingof any ofthe rest,and consequentlythe probabilitythat
the 2nd event happens on suppositionthe 1st does is the same withits originalprobability;
but the probabilitythat any two events happen is a ratio compounded of the probability
of the 1st event,and the probabilityof the 2nd on suppositionthe 1sthappens by prop. 3.
Whereforethe probabilitythat any two independentevents both happen is a ratio com-
pounded of the probabilityof the 1st and the probabilityof the 2nd. And in like manner
consideringthe 1st and 2nd events together as one event; the probability that three
independentevents all happen is a ratio compounded of the probabilitythat the two 1st
both happen and the probabilityof the 3rd. And thus you may proceed iftherebe ever so
many such events; fromwhence the propositionis manifest.
COR. 1. If therebe several independentevents,the probabilitythat the 1st happens the
2nd fails,the 3rdfailsand the 4th happens, etc. is a ratio compoundedof the probabilityof
the 1st,and the probabilityofthe failureofthe 2nd, and the probabilityofthe failureofthe
3rd,and the probabilityofthe 4th,etc. For the failureofan eventmay always be considered
as the happeningof its contrary.
COR. 2. If therebe several independentevents,and the probabilityof each one be a, and

that ofits failingbe b,the probabilitythat the 1sthappens and the 2nd fails,and the 3rdfails
and the 4th happens, etc. will be abba, etc. For, accordingto the algebraic way ofnotation,
ifa denoteany ratio and b another,abba denotesthe ratio compoundedofthe ratiosa, b,b,a.
This corollarythereforeis only a particular case of the foregoing.
DEFINITION. If in consequence of certain data therearises a probabilitythat a certain
event should happen, its happeningor failing,in consequence of these data, I call it's hap-
pening or failingin the 1st trial. And if the same data be again repeated,the happeningor
failingoftheeventin consequence ofthemI call its happeningor failingin the 2nd trial; and
so on as oftenas the same data are repeated. And hence it is manifestthat the happeningor
failingofthe same eventin so manydiffe[rent] trials,is in realitythe happeningor failingof
so many distinctindependentevents exactly similarto each other.
* What is provedby MrBayes in thisand theprecedingpropositionis the samewiththeanswerto the
followingquestion.What is the probabilitythat a certainevent,whenit happens,will be accompanied
withanotherto be determinedat the same time? In this case, as one ofthe eventsis given,nothingcan
be due forthe expectationof it; and, consequently,the value of an expectationdependingon the hap-
peningofboth eventsmustbe the same withthe value ofan expectationdependingon the happeningof
one ofthem. In otherwords;the probabilitythat,whenone oftwo eventshappens,the otherwill,is the
same withthe probabilityof this other. Call x then the probabilitydfthis other,and ifb/Nbe the pro-
bability of the given event, and p/N the probabilityofboth,because p/N = (b/N)x x, x = p/b= the
probabilitymentionedin thesepropositions.

Prop. 7
If the probabilityof an event be a, and that of its failurebe b in each single trial,the
probabilityof its happeningp times,and failingq times in p + q trials is EaPbq if E be the
coefficient of the termin which occurs aPbqwhen the binomial (a + b)P+qis expanded.
For the happeningorfailingofan eventin different trialsare so manyindependentevents.
Wherefore(by cor. 2 prop. 6) the probabilitythat the event happens the 1st trial,failsthe
2nd and 3rd, and happens the 4th, fails the 5th, etc. (thus happening and failingtill the
numberoftimesit happens be p and the numberit failsbe q) is abbabetc. till the numberof
a's bep and thenumberof b's be q, that is; 'tis aPb2. In like mannerifyou considerthe event
as happeningp timesand failingq timesin any otherparticularorder,the probabilityforit is
aPbY;but the numberofdifferent ordersaccordingto whichan event may happen or fail,so
as in ali to happen p times and fail q, in p + q trialsis equal to the numberof permutations
that aaaa bbbadmit of when the numberof a's is p, and the number of b's is q. And this
number is equal to E, the coefficientof the term in which occurs aPbq when (a+ b)P+q is
expanded. The eventthereforemay happen p timesand failq in p + q trialsE different ways
and no more,and its happeningand failingthese several different ways are so many incon-
sistentevents,the probabilityforeach ofwhichis aPb2,and thereforeby prop. 1 the proba-
bilitythat some way or otherit happens p times and fails q times in p + q trialsis EaPb2.
SECTIONII
POSTULATE. 1. I suppose the square table or plane ABCD to be so made and levelled,that if
eitherofthe balls o or W be thrownupon it, thereshall be the same probabilitythat it rests
upon any one equal part ofthe plane as another,and that it mustnecessarilyrestsomewhere
upon it.
2. I suppose that the ball W shall be firstthrown,aid throughthe point where it rests
a line os shall be drawn parallel to AD, and meetingCD and AB in s and o; and'that after-
wardstheball 0 shall be thrownp+ q orn times,and that its restingbetweenAD and os after
a singlethrowbe called the happeningofthe eventM in a singletrial.These thingssupposed:
LEM. 1. The probability that the point o will fall

between any two points in the line AB is the ratio of the c F sH I K *L D
distance betweenthe two points to the whole line AB.
Let any two pointsbe named,asf and b in theline AB,
and throughthem parallel to AD draw fF, bL meeting
CD in F and L. Then if the rectanglesCf, Fb, LA are
commensurableto each other,theymay each be divided
into the same equal parts, which being done, and the
ball W thrown,the probabilityit wili rest somewhere
upon any numberof these equal parts will be the sum of
the probabilitiesit has to rest upon each one of them, B o e d c b
because its restingupon any differentparts of the plane g . t
AC are so many inconsistent events; and this sum, m o.

because theprobabilityit should restupon any one equal o h k
part as anotheris the same, is the probabilityit should
restupon any one equal part multipliedby thenumberof

THOMAS BAYES 303
parts.Consequently, theprobability thereis thattheball Wshouldrestsomewhere uponFb
istheprobability it has to restupononeequal partmultiplied bythenumberofequal parts
inFb; andtheprobability itrestssomewhere uponCforLA, i.e.thatitdoesnotrestuponFb
(becauseit mustrestsomewhere uponAC) is theprobability it restsuponone equal part
multiplied bythenumberofequalpartsinCf,LA takentogether. Wherefore, theprobability
it restsupon Fb is to theprobability it does not as thenumber of equal parts in Fb is to the
numberofequal partsin Cf,LA together, or as Fb to Cf,LA together, or as fb to Bf,Ab
together. Wherefore theprobability it restsuponFb is to theprobability it doesnotas fb to
Bf,Ab together.And (componendo inverse) theprobability it restsuponFb is to theproba-
bilityitrestsuponFb addedtotheprobability it doesnot,asfbto AB, oras theratiooffbto
AB totheratioofAB toAB. But theprobability ofanyeventaddedtotheprobability ofits
failureis theratioofequality;wherefore, theprobability it restsuponFb is to the ratioof
equalityas theratiooffbtoAB to theratioofAB to AB, ortheratioofequality;andthere-
foretheprobability it restsuponFb is theratiooffb to AB. But exhypothesi accordingas
theball WfallsuponFb ornotthepointo willlie betweenf and b ornot, and therefore the
probability thepointo willlie betweenf and b is theratiooffb to AB.
Again; if the rectanglesCf, Fb, LA are not commensurable, yet the last mentioned
probability can be neithergreaternorlessthantheratiooffbto AB; for,ifit be less,letit
be theratiooffctoAB, anduponthelinefbtakethepointsp andt,so thatptshallbe greater
thanfe,and thethreelinesBp, pt,tA commensurable (whichit is evidentmaybe always
donebydividingAB intoequal partslessthanhalfcb,and takingp andtthenearestpoints
ofdivisiontofandc thatlieuponfb).ThenbecauseBp,pt,tAarecommensurable, so arethe
rectanglesCp, Dt, and that upon pt compleating the square AB. Wherefore, by whathas
beensaid,theprobability thatthepointo willlie betweenp and t is theratioofpt to AB.
But ifitliesbetweenp and tit mustliebetweenfand b.Wherefore, theprobability it should
lie betweenfand b cannotbe lessthantheratioofpt to AB, and therefore mustbe greater
thantheratiooffcto AB (sinceptis greaterthanfc). Andafterthesamemanneryoumay
provethattheforementioned probability cannotbe greaterthantheratiooffb to AB, it
musttherefore be thesame.
LEM. 2. The ball W havingbeenthrown,and the line os drawn,theprobability of the
eventM in a singletrialis theratioofAo to AB.
For,in thesamemanneras in theforegoing lemma,theprobability thattheball o being
thrownshallrestsomewhere uponDo orbetween AD and so is the ratioof Ao toAB. Butthe
restingoftheball o betweenAD and so aftera singlethrowis thehappening oftheeventM
in a singletrial.Wherefore thelemmais manifest.
Prop. 8
If uponBA youerectthefigure BghikmA whoseproperty is this,that(thebase BA being
dividedintoanytwoparts,as Ab, and Bb and at thepoint ofdivision ba perpendicularbeing
erectedand terminated by thefigurein m; and y,x, r representing respectivelytheratioof
bm,Ab,and Bb to AB, and E beingthecoefficient oftheterminwhichoccursaPbqwhenthe
binomial(a + b)p+qis expanded)y = ExPrq.I saythatbeforetheball Wis thrown, theproba-
and
bilitythepointo shouldfallbetweenf b,anytwopoints named in thelineAB, and withall
thattheeventM shouldhappenp timesand failq inp + q trials,is theratiooffhilkmb, the
partofthefigure BghikmAintercepted betweentheperpendiculars fy,bmraiseduponthe
lineAB, to CA thesquareuponAB.

304 and statistics.IX
Studiesin thehistoryofprobability
DEMONSTRATION
For ifnot; firstlet it be the ratio ofD a figuregreaterthanfghikmbto CA, and throughthe
points e, d, c draw perpendicularstofb meetingthe curveAmigB in h, i, k; the point d being
so placed that di shall be the longestofthe perpendicularsterminatedby the linefb,and the
curve AmigB; and the points e, d, c being so many and so placed that the rectangles,bk,ci,
ei,fh taken togethershall differless fromfghikmbthan D does; all whichmay be easily done
by the help ofthe equation ofthe curve,and the difference betweenD and the figurefghiklmb
given. Then sincedi is the longestofthe perpendicularordinatesthat insistuponfb, the rest
will graduallydecrease as they are fartherand fartherfromit on each side, as appears from
the constructionof the figure,and consequentlyeh is greaterthan gf or any otherordinate
that insistsupon ef.
Now ifAo were equal to Ae, then by lem. 2 the probabilityofthe event M in a singletrial
would be the ratio of Ae to AB, and consequentlyby cor. Prop. 1 the probabilityof it's
failurewould be the ratio ofBe to AB. Wherefore,ifx and r be the two forementionedratios
respectively,by Prop. 7 the probabilityof the event M happeningp times and failingq in
p + q trialswould be ExPel. But x and r being respectivelythe ratios ofAe to AB and Be to
AB, ifyis the ratio ofehto AB, then,by constructionofthe figureAiB,y = ExPr . Wherefore,
if Ao were equal to Ae the probabilityof the event M happeningp times and failingq in
p + q trialswould be y,orthe ratio of ehto AB. And if Ao wereequal to Af,or were any mean
betweenAe and Af,the last mentionedprobabilityforthe same reasonswould be the ratio of
fg or some otherof the ordinatesinsistingupon ef,to AB. But eh is the greatestof all the
ordinates that insist upon ef. Wherefore,upon suppositionthe point should lie anywhere
betweenf and e, the probabilitythat the event M happens p times and failsq in p + q trials
cannot be greaterthan the ratio ofehto AB. There thenbeingthese two subsequent events,
the 1st that the point o will lie between e and f, the 2nd that the event Ml will happen
p times and fail q in p + q trials, and the probabilityof the first(by lemma 1) is the ratio
of ef to AB, and upon supposition the 1st happens, by what has been now proved, the
probabilityofthe 2nd cannot be greaterthan the ratio ofehto AB, it evidentlyfollows(from
Prop. 3) that the probabilityboth togetherwill happen cannot be greaterthan the ratio
compounded of that of efto AB and that of eh to AB, which compoundratio is the ratio of
fh to CA. Wherefore,the probabilitythat the point o will lie betweenf and e, and the event
M happen p times and fail q, is not greaterthan the ratio offh to CA. And in like manner
theprobabilitythe pointo willlie betweene and d, and the eventM happen and failas before,
cannot be greaterthan the ratio of ei to CA. And again, the probabilitythe point o will lie
between d and c, and the event M happen and fail as before,cannot be greaterthan the
ratio ofci to CA. And lastly,the probabilitythat the pointo willlie betweenc and b,and the
event M happen and fail as before,cannot be greaterthan the ratio of bkto CA. Add now
all these several probabilitiestogether,and theirsum (by Prop. 1) will be the probability
that the point will lie somewherebetweenf and b, and the event M happen p times and fail
q in p + q trials. Add likewisethe correspondentratios together,and theirsum will be the
ratio ofthe sum ofthe antecedentsto theircommonconsequent,i.e. the ratio offh,ei, ci, bk
togetherto CA; whichratiois less than that ofD to CA, because D is greaterthanfh,ei, ci, bk
together.And therefore, the probabilitythat the point o will lie betweenfand b,and withal
that the eventM willhappen p timesand failq inp + q trials,is less than theratioofD to CA;
but it was supposed the same whichis absurd. And in like manner,by inscribingrectangles

THOMAS BAYES 305
withinthe figure,
as eg,dh,dk,cm,you mayprovethatthe last mentioned
probability
is
greater
thantheratioofanyfigure lessthanfghikmbto CA.
Wherefore,thatprobabilitymustbe theratiooffghikmb to CA.
COR. Beforethe ball W is thrownthe probability
that the pointo willlie somewhere
A
between andB, orsomewhere uponthelineAB, and withalthattheeventM willhappen
p times,and failq inp + q trialsis theratioofthewholefigure
AiB to CA. But it is certain
thatthepointo willlie somewhere uponAB. Wherefore,beforethe ball W is thrownthe
theeventA willhappenp timesandfailq inp + q trialsis theratioofAiB to CA.
probability
Prop. 9
If beforeanythingis discoveredconcerningtheplaceofthepointo,it shouldappearthat
theeventM hadhappenedp timesandfailedq inp + q trials,andfromhenceI guessthatthe
pointo lies betweenany two pointsin the lineAB, as f and b,and consequently thatthe
probability oftheeventM ina singletrialwassomewhere betweentheratioofAbtoAB and
thatofAfto AB: theprobability I am in therightis theratioofthatpartofthefigure AiB
describedas beforewhichis intercepted betweenperpendiculars erecteduponAB at the
pointsfand b,to thewholefigure AiB.
For,therebeingthesetwosubsequentevents,thefirstthatthepointo willlie between
f and b; thesecondthattheeventM shouldhappenp timesandfailq inp + q triAls;and (by
cor.prop.8) theoriginalprobabilityofthesecondis theratioofAiB to CA,and (byprop.8)
the probability ofbothis the ratiooffghimb to CA; wherefore (by prop.5) it beingfirst
discovered thatthesecondhashappened,andfromhenceI guessthatthefirst hashappened
also,theprobabilityI am in therightis theratiooffghimb to AiB, the pointwhichwas to
be proved.
COR.The samethingssupposed,ifI guessthattheprobability
oftheeventM liessome-
wherebetween0 and the ratioofAb to AB, mychanceto be in the rightis the ratioof
Abmto AiB.
Scholium
Fromthepreceding proposition itis plain,thatinthecaseofsuchan eventas I therecallM,
fromthenumberoftimesithappensandfailsina certainnumberoftrials,withoutknowing
anything moreconcerning it,onemaygivea guesswhereabouts it's probability
is, and,by
theusual methodscomputing themagnitudes oftheareastherementioned, see thechance
thattheguessis right.Andthatthesameruleis theproperoneto be usedin thecase ofan
eventconcerning theprobability ofwhichwe absolutelyknownothingantecedently to any
trialsmade concerning it, seems to appear from the followingconsideration; viz. that
concerning suchan eventI have no reasonto thinkthat,in a certainnumberoftrials,it
shouldratherhappenanyonepossiblenumberoftimesthananother.For,on thisaccount,
I mayjustlyreasonconcerning it as ifits probability
had been at firstunfixed, and then
determined in sucha manneras to givemeno reasonto thinkthat,in a certainnumberof
trials,it shouldratherhappenanyonepossiblenumberoftimesthananother.But thisis
exactlythe case ofthe eventM. For beforethe ball W is thrown,whichdetermines it's
probability ina singletrial(bycor.prop.8),theprobability ithasto happenp timesandfail
q inp + q orn trialsis theratioofAiB to CA, whichratiois thesamewhenp + q orn is given,
whatevernumberp is; as willappearby computing themagnitudeofAiB by themethod

306 Studies in the historyofprobabilityand statistics. IX
* Andconsequently
offluxions. beforetheplaceofthepointo is discovered orthenumberof
times theevent M has happened in n I
trials, can have no reasonto think it shouldrather
happenone possiblenumberoftimesthananother.
In whatfollowstherefore I shalltakeforgrantedthattherulegivenconcerning theevent
M in prop.9 is also theruleto be usedinrelationto anyeventconcerning theprobability of
which nothing at all is known antecedently to any trialsmade or observed concerning it.
Andsuchan eventI shallcall an unknownevent.
COR. Hence,bysupposing theordinatesin thefigure AiB to be contracted in theratioof
E to one,whichmakesno alterationin theproportion of theparts ofthefigure intercepted
betweenthem,and applyingwhatis said oftheeventM to an unknownevent,we havethe
followingproposition, whichgivestherulesforfinding theprobability ofan eventfromthe
numberoftimesit actuallyhappensand fails.
Prop. 10
be describeduponany base AH (Vid. Fig.) havingforit's equationy = xPrq;
If a figure
wherey,x, r arerespectively theratiosofan ordinateofthefigure on thebase at
insisting
rightangles,ofthesegment ofthe base intercepted between the
ordinateand A the beginning ofthe base, and oftheotherseg-
mentofthebase lyingbetweentheordinateand thepointH, to
the base as theircommonconsequent.I say then that if an
unknowneventhas happenedp timesand failedq inp + q trials,
and in the base AH takingany two pointsas f and t you erect F D
theordinatesfO,tF at rightangleswithit,thechancethatthe C
probability oftheeventlies somewhere betweentheratioofAf H h A
to AH and that ofAt to AH, is the ratiooftFCf,thatpartof t f
thebefore-described figurewhichis intercepted betweenthetwo
ordinates,to ACFH thewholefigure insisting on thebase AH.
Thisisevidentfromprop.9 andtheremarks madeintheforegoing scholiumand corollary.
Now,in orderto reduce the foregoing rule to practice,we mustfindthe value of the
areaofthefigure describedand theseveralpartsofit separated,byordinatesperpendicular
to itsbase. For whichpurpose,supposeAHf- 1 and HO thesquareuponAH likewise= 1,
and Cfwillbe = y,and Af = x, and Hf = r,becausey,x and r denotetheratiosofCf,Af,
and Hf respectively to AH. And by the equation of the curvey xPrq and (because
Af+ fH- AH) r+ x = 1. Wherefore
y = XP(I _x)q
-
= PqxP+l +q(q 1)xP+S q(q -1)(q -2)xp +etc.
2 2.3 +ec
Now the abscisse being x and the ordinate xP the correspondentarea is xP+l/(p+ 1) (by
prop. 10, cas. 1, Quadrat. Newt.)t and the ordinatebeing qxP+1the area is qxP+2/(p+ 2); and
* It willbe provedpresentlyinart. 4 by computingin themethodherementionedthatAiB contracted
in the ratio of E to 1 is to CA as 1 to (n + 1)E: fromwhence it plainlyfollowsthat,antecedentlyto this
contraction,AiB must be to CA in the ratio of 1 to n + 1, which is a constantratio when n is given,
whateverp is.
t 'Tis veryevidenthere,withouthaving recourseto Sir Isaac Newton,that the fluxionof the area
ACf being q(q-1)
yx = xPx-qxP+li++ 2 x"+x-etc.
xP+1 qx P+2 q(q-1) xV+3

the fluentor area itselfis p 2+ et
p+1 p+2 2(p +3)

THOMAS BAYES 307
in like mannerof the rest. Wherefore,
the abscisse being x and the ordinatey or
xP - qxl+1 + etc. the correspondent area is
XP+l1 qxP+2 q(q- 1) Xp+3 _q(q- 1) (q-2) xP+ + etc.

p+1 p+2 2(p+3) 2.3(p+4)
Wherefore,if x = Af = Af/(AH), and y = Cf = Cf/(AH), then
ACf XP+l qxP+2 q(q-l)XP+3
AfHO=p?1 p+2 2(p?+3)

From which equation, if q be a small number,it is easy to findthe value of the ratio of
ACf to HO and in like manneras that was found out, it will appear that the ratio of HCf
to HO is rq+l qpr+2 qp(p-l)r+3 p(p- )(p'2) r+4
q+1 q+2 2(q+3) 2.3(q+4)

which serieswill consistof fewtermsand thereforeis to be used whenp is small.
2. The same thingssupposed as before,the ratio of ACf to HO is
xP+lrq qxP+2rq-l q(q - 1) XP+3rq-2 q(q- 1) (q -2) Xp+4rq-3
p+l (p+l)(p+2) (p+ l)(p+2)(p+3) (p+l)(p+2)(p+3)(p+4)
(n + 1) (p + 1) (p + 2)
... n' +
wheren = p + q. For this seriesis the same withxP+1/(p+ 1) - qxP+2/(p+ 2) + etc. set down
etc.+
in Art. 1stas thevalue oftheratioofACf to HO; as willeasilybe seenby puttingin theformer
instead of r its value 1- x, and expanding the terms and orderingthem according to the
powersofx. Or, morereadily,by comparingthe fluxionsofthe two series,and in the former
instead of r substituting-.*
3. In like manner,the ratio of HICfto HO is
rq+lXP
+ prq+2xP-l p(p - 1) rq+3Xp-2
q+1 (q+1)(q+2) (q+1)(q+2)(q+3)

* The fluxionof the firstseriesis
qxp+Lrq l+ qxP+lrq-lx q(q - 1) XP+2rq-2r q(q -1) XP+2rq-2j q(q - 1) (q -2) x2+3rq-3a
xprq+ + + + +ec
p+1 -+ p+1 (p+ l)(p+2) (p+ l)(p+2) (p+ l)(p+2)(p+3)
or, substituting-xi forr,
qxP+1r-lS qxP+lrq-1 q(q - 1) x2+2r7-2X q(q-1) x2+2r-2-e -etc.
xPrx- + - +
p+1 p+1 (p+ l)(p+2) (p+ l)(p+2)
which,as all the termsafterthe firstdestroyone another,is equal to
xPrq = xP(1-x)2 = xP.{ I-qx +q (q 2) X2-etc.
2
=xPx, - qx P+ILx (q_1)p2Xetc.
2 qxv+
q (q -1) X
2
the fluxionof the latterseries,or of
= -ff + etc.
p+1 p+2
The two seriesthereforeare the same.

308 Studie8 in thehistoryof probabilityand stati8tiC8. IX
ofthat termofthe binomial (a + b)p+q expanded in whichoccurs
4. If E be the coefficient
ab , the ratio of the whole figureACFH to HO is {(n + 1) E}', n being = p + q. For, when
Af = AH, x =.1, r = 0. Wherefore,all thetermsof theseriesset down in Art.2 as expressing
the ratio of ACIfto HO will vanish except the last, and that becomes
q(q- 1)...I
(n + 1) (p + 1) (p + 2) ...n
of that termin the binomial (a + b)nexpanded in which occurs
But E being the coefficient
aPb2is equal to (p + 1) (p + 2)...n
q(q- 1)...1
And, because Afis supposed to become = AH, ACf = ACH. From whence this article is
plain.
5. The ratio of ACfto the whole figureACFH is (by Art. 1 and 4)
[XP+l qXp+2 q(q -l) Xp+3 1
+ 1) E
(n (n+1)E
+lp+2 --2(p -H3) _etc.
and if, as x expresses the ratio of Af to AH, X should express the ratio of At to AH; the
ratio of AFt to ACFFHwould be
X+1 qXP+2 q(q -1) Xp+3 _
(n+i1)E - + etc.
and consequentlythe ratio of tFCfto ACFH is (n+ 1) E multipliedinto the difference

betweenthe two series. Comparethiswithprop. 10 and we shall have the followingpractical
rule.
Rule 1
If nothingis knownconcerningan event but that it has happened p timesand failedq in
p + q or n trials,and fromhence I guess that the probabilityofits happeningin a singletrial
lies somewherebetween any two degreesof probabilityas X and x, the chance I am in the
rightin my guess is (n + 1) E multipliedinto the difference betweenthe series
XP+' qXP+2 q(q-1)Xp+3
p+l p+2 + 2(p+3) etc
* ~~~~Xp+1qXp+2 q(q -1) Xp+3

and the series P+ qxp+2 q(q - 1) -etc.
p + 1p+2 2(p +3)
E being the coefficientof aPbqwhen (a + b)nis expanded.
This is the properrule to be used when q is a small number; but if q is large and p small,
change everywherein the serieshere set down p into q and q into p and x into r or (1 -x),
and X into R = (1 - X); which will not make any alterationin the differencebetween the
two series.
Thus farMr Bayes's essay.
With respectto the ruleheregiven,it is furtherto be observed,that whenbothp and q are
verylargenumbers,it willnot be possibleto apply it to practiceon account ofthe multitude
of termswhichthe seriesin it willcontain. Mr Bayes, therefore,by an investigationwhichit
would be too tedious to give here,has dedaced fromthis rule another,whichis as follows.

THOMAS BAYES 309
Rule 2
If nothingis knownconcerningan event but that it has happened p times and failed q in
p + q or n trials,and fromhence I guess that the probabilityofits happeningin a singletrial
lies between (p/n) + z and (p/n) - z; if m2 = n3/(pq), a = p/n, b = q/n,E the coefficientof the
termin whichoccurs aPbqwhen (a + b)n is expanded, and
(n+ V(2pq)
l) EaPbq
nvn
multipliedby the series
m3z3 (n.-2) m5z5 (n-2) (n-4) m7z7 (n-2) (n-4) (n-6) m9z9
m_ + + ec
3 2n.5 2n.3n.7 2n.3n.4n.9
my chance to be in the rightis greaterthan
2 *
1 + 2Eapbq + 2Eapbq/n
and less than

1- 2Eabq -2EaPbq/n'
and ifp = q my chance is 2X exactly.

In orderto renderthis rule fitforuse in all cases it is only necessaryto know how to find
withinsufficient nearnessthe value of EaPbq and also of the series mz -_ m3z3 + etc. With
respect to the formerMr Bayes has proved that, supposing K to signifythe ratio of the
quadrantal arc to its radius, EaPbq will be equal to Vn/I(Kpq) multiplied by the ratio,
[h], whose hyberboliclogarithmis
1rl1111 1 [1 1 1]
12 n p q 360 n3 p3 q3
1 [11 11 1 -11 1] 1 [1 111

+ --- I I- - -I+ etc.---
1260 n5 p5 q5- 1680[n7 p7q7 q 1188 n9 p9 q9j-etc.t
* In Mr Bayes's manuscriptthis chance is made to be greaterthan 2z/(1 + 2EaPb2) and less than
2 -/( -2EaPbq). The thirdtermin the two divisors,as I have giventhem,beingomitted.But thisbeing
evidentlyowingto a small oversightin the deductionofthisrule,whichI have reasonto thinkMr Bayes
had himselfdiscovered,I have venturedto correcthis copy, and to give the rule as I am satisfiedit
oughtto be given.
t A veryfewtermsofthisserieswill generallygive the hyperboliclogarithmto a sufficient degreeof
exactness. A similarseries has been given by Mr DeMoivre, Mr Simpson and other eminentmathe-
maticiansin an expressionforthe sum ofthelogarithmsofthenumbers1, 2, 3, 4, 5, to x, whichsum they
have assertedto be equal to
alogc+(x+ -ff
)logx-x+ - + -etc.
~ 12x 360X3 1260X5
c denotingthe circumference ofa circlewhoseradiusis unity. But MrBayes, in a precedingpaper in this

volume, has demonstratedthat, thoughthis expressionwill very nearlyapproach to the value of this
sum whenonlya propernumberofthefirsttermsis taken,the wholeseriescannotexpressany quantity
at all, because, let x be what it will,therewillalways be a part ofthe serieswhereit willbeginto diverge.
This observation,thoughit does not much affectthe use of this series,seems well worththe notice of
mathematicians.
20 Biom. 45

where the numeral coefficientsmay be found in the following manner. Call them
A, B, C,D, E etc. Then
1 I1 1 A 1 lOB+A
A= =- B= IC- - __
2.2.3 3.4' 2.4.5 3' - 2.6.7 5
1 35C+21B+A 1 126C + 84D + 36B + A
2.8.9 7 ' 2.10.11 9
1 462D + 330C + 165E +?55B + A
2.12.13 11 etc.
wherethe coefficients of B, C, D, E, F, etc. in the values of D, E, F, etc. are the 2, 3, 4, etc.

highestcoefficients in (a + b)7, (a + b)9,(a + b)1, etc. expanded; affixingin every particular
value the least ofthese coefficients to B, the next in magnitudeto the furthestletterfromB,
the next to C, the next to the furthestbut one, the next to D, the next to the furthestbut
two, and so on.*
With respectto the value of the series
mz - im3z3 +(n -2)rm5Z5 etc.
he has observedthat it may be calculated directlywhen mz is less than 1,or even not greater
than j3: but when mz is much largerit becomes impracticableto do this; in which case he
shews a way of easily findingtwo values of it verynearlyequal betweenwhichits truevalue
must lie.
The theoremhe gives forthis purpose is as follows.
Let K, as before,stand forthe ratio ofthe quadrantal arc to its radius, and H forthe ratio
whose hyperboliclogarithmis
22 -1 24 -1 26 _1 28 -
2n 360n3+ 1260n5 1680n7+etc

Then the series mz- M3z3+ etc. will be greateror less than the 'series
Hn 1K n(- n) n2 n- )
(n+ 1)12 (n+ 2)2mz (n+ 2) (n+ 4) 4m3z3

3n(I 2r2z2 jn+3 3.5. n4(1 2M2z2) Jn+4
(n + 2) (n+ 4) (n+ 6)8m5z5+ (n - 2) (n+ 4) (n+ 6) (n+ 8) 16m7z7

continuedto any numberofterms,accordingas thelast termhas a positiveor a negative
signbeforeit.
FromsubstitutingthesevaluesofEaPbqand
mz- m3z3 (n-2)" m5z5 etc.

3 ?+ 2n 5
in the secondrulearisesa thirdrule,whichis theruleto be used whenmzis ofsomecon-
siderablemagnitude.
* This method of findingthese coefficientsI have deduced from the demonstration of the third lemma
at the end of Mr Simpson's Treatise on the Nature ond Laws of Chance.

THOMAS BAYES 311
Rule 3
If nothingis knownof an event but that it has happened p times and failed q in p + q or
n trials, and fromhence I judge that the probabilityof its happeningin a singletrial lies
betweenp/n+ z and p/n- z my chance to be rightis greaterthan
V(Kpq) h J2H 12 (n + 1) (1- 2M2Z2/n)in+l1

4(Kpq) + hni + hn-A VK (n + 2) mz
and less than
2 <(Kpq) h {2H _2 (n + 1) (1 - 2m2z2/n)in+l 12 n (n + 1) (1 - 2M2z2/n)in+2

V(Kpq) - hni - hn 1 (n + 2) mz
4K VK (n + 2) (n + 4) 2m3z3
where in2,K, h and Hfstand forthe quantities already explained.
AN APPENDIX
Containing an application of the foregoing Rules to some particular Cases

The firstrulegivesa directand perfectsolutionin all cases; and thetwofollowingrulesare onlyparticular
methodsofapproximatingto the solutiongivenin the firstrule,whenthe labour ofapplyingit becomes
too great.
The firstrulemay be used in all cases whereeitherp or q are nothingornot large.The secondrulemay
be used in all cases wheremzis less thanJ3;and thethirdin all cases whereM2z2 is greaterthan 1and less
than -2n,ifn is an even numberand verylarge. If n is not large this last rule cannot be much wanted,
because, m decreasingcontinuallyas n is diminished,the value ofz may in thiscase be taken large,(and
therefore a considerableintervalhad betweenp/n- z and p/n+ z), and yetthe operationbe carriedon by
the second rule; or mz not exceed V3.
But in orderto shewdistinctlyand fullythe natureof thepresentproblem,and how farMr Bayes has
carriedthe solutionofit; I shall give the resultofthissolutionin a fewcases, beginningwiththe lowest
and most simple.
Let us thenfirstsuppose,of such an eventas that called M in the essay,or an eventabout the proba-
bilityofwhich,antecedentlyto trials,we knownothing,thatit has happenedonce,and thatit is enquired
what conclusionwe may draw fromhencewithrespectto the probabilityof it's happeningon a second
trial.
The answeris thattherewouldbe an odds ofthreeto one forsomewhatmorethanan even chancethat
it would happen on a second trial.
For in this case, and in all otherswhereq is nothing,the expression
(XP+1 XV+'
p+1
(n + 1) -p+1 or XP+1-xP+1
gives the solution, as will appear fromconsideringthe firstrule. Put thereforein this expression
p +l1 = 2, X = I and x = 2 and it willbe 1- (2)2 or 3; whichshewsthechancethereis thattheprobability
ofan eventthathas happenedonce lies somewherebetween1 and 1; or (whichis the same) theodds that
it is somewhatmorethan an even chance that it will happen on a second trial.\*
In thesame manneritwillappear thatiftheeventhas happenedtwice,theodds nowmentionedwillbe
seven to one; if thrice,fifteento one; and in general,if the event has happened p times,therewill be
an odds of 2P+1 - 1 to one, formore than an equal chance that it will happen on furthertrials.
Again, suppose all I know of an event to be that it has happened ten timeswithoutfailing,and the
enquiryto be what reason we shall have to thinkwe are rightif we guess that the probabilityof it's
happeningin a singletrial lies somewherebetween 17 and 2, or that the ratio of the causes of it's
happeningto those of it's failureis some ratio betweenthat of sixteento one and two to one.
Herep + 1 = 11,X = 4-4 andx = 2 and XP+1 -xP+1 = (!- )" - (.3)11 = 0.5013 etc. The answertherefore
is, that we shall have verynearlyan equal chance forbeing right.
* There can, I suppose,be no reasonforobservingthat on thissubject unityis always made to stand
forcertainty,and i foran even chance.

and statistics.IX
In thismannerwe may determinein any case what conclusionwe oughtto draw froma givennumber
of experimentswhichare unopposed by contraryexperiments.Every one sees in generalthat thereis
reasonto expectan eventwithmoreor less confidenceaccordingto the greateror less numberoftimesin
which,under given circumstances,it has happened withoutfailing;but we here see exactly what this
reason is, on what principlesit is founded,and how we oughtto regulateour expectations.
But it will be properto dwell longeron thishead.
Suppose a solid or die ofwhosenumberofsides and constitutionwe knownothing;and that we are to
judge of thesefromexperimentsmade in throwingit.
In this case, it should be observed,that it would be in the highestdegreeimprobablethat the solid
should, in the firsttrial,turnany one side which could be assigned beforehand;because it would be
known that some side it must turn,and that therewas an infinityof other sides, or sides otherwise
marked,whichit was equally likelythat it should turn. The firstthrowonly shews that it has the side
thenthrown,withoutgivingany reasonto thinkthat it has it any one numberoftimesratherthan any
other.It willappear,therefore, thatafterthefirstthrowand notbefore,we shouldbe in thecircumstances
requiredby the conditionsof the presentproblem,and that the whole effectof this throwwould be to
bringus intothesecircumstances.That is: theturningtheside firstthrownin any subsequentsingletrial
would be an event about the probabilityor improbabilityofwhichwe could formno judgment,and of
whichwe should know no more than that it lay somewherebetweennothingand certainty.With the
second trialthenour calculationsmustbegin; and ifin that trialthe supposedsolid turnsagain the same
side,therewillarisethe probabilityofthreeto one that it has moreofthat sortofsides thanofall others;
or (whichcomes to the same) that thereis somewhatin its constitutiondisposingit to turnthat side
oftenest:And thisprobabilitywillincrease,in themanneralreadyexplained,withthenumberoftimesin
whichthat side has been thrownwithoutfailing.It shouldnot,however,be imaginedthat any iumber
ofsuch experimentscan give sufficient reason forthinkingthat it would neverturnany otherside. For,
suppose it has turnedthe same side in everytriala millionoftimes. In thesecircumstancestherewould
be an improbabilitythat it has less than 1,400,000more of these sides than all others;but therewould
also be an improbabilitythat it had above 1,600,000timesmore.The chance forthe latteris expressed
by 1,600,000/1,600,001 raisedto the millionethpowersubtractedfromunity,whichis equal to 0-4647etc
and the chance forthe formeris equal to 1,400,000/1,400,001 raised to the same power,or to 0-4895;
which,being both less than an equal chance, proves what I have said. But thoughit would be thus
improbablethatit had above1,600,000timesmoreor lessthan 1,400,000timesmoreofthesesidesthan of
all others,it by no meansfollowsthatwe have any reasonforjudgingthat thetrueproportionin thiscase
lies somewherebetweenthat of 1,600,000to one and 1,400,000to one. For he that willtake the pains to
make the calculationwillfindthat thereis nearlythe probabilityexpressedby 0-527,or but littlemore
than an equal chance,that it lies somewherebetween that of 600,000to one and threemillionsto one.
It may deserveto be added, that it is moreprobable that this proportionlies somewherebetweenthat
of 900,000to 1 and 1,900,000to 1 than betweenany othertwo proportionswhoseantecedentsare to one
anotheras 900,000to 1,900,000,and consequentsunity.
I have made these observationschieflybecause they are all strictlyapplicable to the events and
appearances ofnature. Antecedentlyto all experience,it would be improbableas infiniteto one, that
any particularevent, beforehandimagined,should followthe applicationof any one natural object to
another;because therewouldbe an equal chanceforany one ofan infinity of otherevents. But ifwe had
once seen any particulareffects, as theburningof wood on puttingit into fire,or the fallingof a stone
on detachingit fromall contiguousobjects, thenthe conclusionsto be drawnfromany numberof sub-
sequent eventsofthe same kindwouldbe to be determinedin the same mannerwiththe conclusionsjust
mentionedrelatingto the constitutionofthe solid I have supposed. In otherwords.The firstexperiment
supposed to be ever made on any natural object would only informus of one event that may follow
a particularchange in the circumstancesof those objects; but it would not suggestto us any ideas of
uniformity in nature,orgiveus theleast reasonto apprehendthat it was, in thatinstanceor in any other,
regularratherthanirregularin its operations.But ifthesame eventhas followedwithoutinterruptioin in
any one ormoresubsequentexperiments,thensome degreeofuniformity willbe observed;reasonwillbe
givento expect the same successin furtherexperiments,and the calculationsdirectedby the solutionof
this problemmay be made.
One example here it will not be amiss to give.
Let us imagineto ourselvesthe case ofa personjust broughtforthinto thisworld,and leftto collect
fromhis observationofthe orderand courseofeventswhat powersand causes take place in it. The Sun
would,probably,be the firstobject thatwould engagehis attention;but afterlosingit the firstnighthe
would be entirelyignorantwhetherhe shouldeversee it again. He would therefore be in the conditionof
a personmakinga firstexperimentabout an event entirelyunknownto him. But let him see a second
appearance or one returnofthe Sun,and an expectationwould be raisedin himofa secondreturn,and he

THOMAS BAYES 313
mightknow that therewas an odds of 3 to 1 forsomeprobabilityof this. This odds would increase,as
beforerepresented,withthenumberofreturnsto whichhe was witness.But no finitenumberofreturns
would be sufficient to produceabsolute or physicalcertainty.For let it be supposed that he has seen it
returnat regularand stated intervalsa millionof times.The conclusionsthis would warrantwould be
suchas follow.Therewouldbe theodds ofthemillionethpowerof2,to one,that itwas likelythatitwould
returnagain at the end ofthe usual interval.There would be the probabilityexpressedby 0-5352,that
the odds forthiswas notgreaterthan 1,600,000to 1; and theprobabilityexpressedby 0-5105,that it was
not less than 1,400,000to 1.
It shouldbe carefullyrememberedthat thesedeductionssuppose a previoustotal ignoranceofnature.
Afterhavingobservedforsometimethecourseofeventsit would be foundthat the operationsofnature
are in generalregular,and that the powersand laws whichprevail in it are stable and permanent.The
considerationofthiswillcause one or a fewexperimentsoftento producea muchstrongerexpectationof
successin furtherexperimentsthanwould otherwisehave been reasonable;just as the frequentobserva-
tion that thingsofa sortare disposedtogetherin any place would lead us to conclude,upon discovering
thereany object of a particularsort,that thereare laid up withit many othersof the same sort. It is
obviousthatthis,so farfromcontradictingtheforegoing deductions,is onlyone particularcase to which
theyare to be applied.
What has been said seemssufficient to shew us what conclusionsto drawfromuniformexperience. It
demonstrates,particularly,that insteadofprovingthat eventswillalwayshappen agreeablyto it,there
will be always reason against this conclusion. In otherwords,wherethe courseof naturehas been the
mostconstant,we can have onlyreasonto reckonupon a recurrency ofeventsproportionedto thedegree
of this constancy;but we can have no reason forthinkingthat thereare no causes in naturewhichwill
everinterfere withthe operationsofthe causes fromwhichthisconstancyis derived,or no circumstances
ofthe worldin whichit willfail. And if this is true,supposingour only data derived fromexperience,
we shall findadditionalreason forthinkingthus if we apply otherprinciples,or have recourseto such
considerationsas reason,independentlyof experience,can suggest.
But I have gonefurtherthan I intendedhere; and it is timeto turnourthoughtsto anotherbranchof
this subject: I mean, to cases wherean experimenthas sometimessucceeded and sometimesfailed.
Here, again, in orderto be as plain and explicitas possible,it willbe properto put the followingcase,
whichis the easiest and simplestI can thinkof.
Let us thenimaginea personpresentat the drawingofa lottery,who knowsnothingofits schemeor of
theproportionofBlanks to Prizes in it. Let it furtherbe supposed,thathe is obligedto inferthisfromthe
numberof blankshe hears drawn compared with the numberof prizes; and that it is enquired what
conclusionsin these circumstanceshe may reasonablymake.
Let him firsthear tenblanks drawnand oneprize,and let it be enquiredwhat chance he will have for
beingrightifhe guessesthat the proportionofblanksto prizesin the lotterylies somewherebetweenthe
proportionsof 9 to 1 and 11 to 1.
Here taking X = 4, x = 9O ,p = 10,q = 1, n = 11, E = 11, the required chance, according to the first
rule,is (n + 1) E multipliedby the difference between
{XP+1 qxp+28 )12~~~~~~~~~~~~~~~~~~~~~~~x+1

xP+l x
qxp+28 ( 1 1 )11 ( 1 12)12- -(_9
(_9)1 )11 (_9
X qX- qand _ =122_ 11 12_12 0 12 = 0-07699etc.
12j}
-
tp+1 p+2 p+1 p+2 . . 11 12 11
There would thereforebe an odds of about 923 to 76, or nearly 12 to 1 against his being right. Had
he guessedonlyin generalthat therewereless than 9 blanks to a prize,therewould have been a proba-
bilityofhis beingrightequal to 0-6589,or the odds of 65 to 34.
Again,supposethathe has heard 20 blanksdrawnand 2 prizes;what chancewillhe have forbeingright
ifhe makes the same guess?
Here X and x beingthesame,we have n = 22,p = 20,q = 2,E = 231,and therequiredchanceequal to
+
(n+l)E
FXP+1
U -
qXP+2 q(q -1)
+II
XP+3 FX+1 qxP+2 q(q -1) xP+3Tt
+ = 0-10843 etc.
tLp+l p+2 2(p+3) J Lp+l p+2 2(p+3) j)
He will,therefore, have a betterchance forbeingrightthan in the formerinstance,the odds against

himnow being892 to 108 or about 9 to 1. But shouldhe onlyguess in general,as before,that therewere
less than 9 blanks to a prize,his chance forbeingrightwill be worse; forinstead of 0-6589or an odds of
near two to one, it will be 0-584,or an odds of 584 to 415.
Suppose, further,that he has heard 40 blanksdrawn and 4 prizes; what will the before-mentioned
chances be?

and statistics.IX
The answerhereis 0 1525,forthe formerofthesechances; and 0-527,forthe latter.Therewill,there-
fore,now be an odds of only 5~ to 1 against the proportionofblanks to prizeslyingbetween9 to 1 and
11 to 1; and but littlemorethan an equal chance that it is less than 9 to 1.
Once more. Suppose he has heard 100 blanksdrawnand 10 prizes.
The answerheremay stillbe foundby thefirstrule; and the chancefora proportionofblanksto prizes
less than 9 to 1 will be 0-44109,and fora proportiongreaterthan 11 to 1, 0-3082. It would therefore
be likelythat therewerenotfewerthan 9 or morethan 11 blanks to a prize. But at the same timeit will
remainunlikely*thatthetrueproportionshouldlie between9 to 1 and 11 to 1, the chance forthisbeing
0-2506etc. There will thereforebe stillan odds ofnear 3 to 1 against this.
From these calculationsit appears that, in the circumstancesI have supposed,the chance forbeing
rightin guessingtheproportionofblanksto prizesto be nearlythesame withthatofthenumberofblanks
drawnin a giventimeto thenumberofprizesdrawn,is continuallyincreasingas thesenumbersincrease;
and that therefore, when they are considerablylarge,this conclusionmay be looked upon as morally
certain. By parityof reason, it followsuniversally,with respectto every event about which a great
numberofexperimentshas been made, that the causes ofits happeningbear the same proportionto the
causes ofitsfailing,withthenumberofhappeningsto thenumberoffailures;and that,ifan eventwhose
causes are supposedto be known,happens ofteneror seldomerthan is agreeableto thisconclusion,there
willbe reasonto believethat thereare some unknowncauses whichdisturbthe operationsofthe known
ones. With respect,therefore,particularlyto the course of events in nature,it appears, that thereis
demonstrativeevidenceto provethattheyare derivedfrompermanentcauses, or laws originallyestab-
lishedin the constitutionofnaturein orderto producethat orderof eventswhichwe observe,and not
fromany ofthe powersofchance.t This is just as evidentas it would be, in the case I have insistedon,
that the reasonofdrawing10 timesmore blanksthan prizesin millionsof trials,was, that therewerein
the wheelabout so many moreblanksthanprizes.
But to proceeda littlefurtherin the demonstrationof thispoint.
We have seen that supposinga person,ignorantof the whole scheme of a lottery,should be led to
conjecture,fromhearing100 blanksand 10 prizes drawn,that the proportionof blanksto prizes in the
lotterywas somewherebetween9 to 1 and 11 to 1, the chance forhis beingrightwould be 0-2506etc.
Let [us] now enquirewhat thischance would be in some highercases.
Let it be supposed that blankshave been drawn 1000 times,and prizes 100 timesin 1100 trials.
In thiscase the powersofX and x rise so high,and the numberof termsin the two series
XP+1 qXP+2 XP+1 qXP+2
- etc. and - etc.
p+1 p+2 p+1 p+2
become so numerousthat it would requireimmenselabour to obtain the answerby the firstrule. 'Tis
necessary,therefore,to have recourseto the second rule. But in orderto make use of it,the interval
betweenX and x mustbe a littlealtered. 100- 90 is 1-0,and thereforethe intervalbetween" -I and
-1l + 1-f4willbe nearlythe same withthe intervalbetween-9 and "4,onlysomewhatlarger.If thenwe
make thequestionto be; what chancetherewouldbe (supposingno moreknownthan that blanks have
been drawn 1000 timesand prizes 100 timesin 1100trials)that the probabilityof drawinga blank in a
singletrialwouldlie somewherebetween1 1- 1 and 1? + f1 we shallhave a questionofthesame kind
withthe precedingquestions,and deviate but littlefromthe limitsassignedin them.
The answer,accordingto the second rule,is that this chance is greaterthan
2E
2EaPbq
1 + 2EaPb +
n
and less than 2E
1- _
-2E b
n
E being
(n+1) V(2pq)Eb M3Z3 (n -2) M5Z5
(~~mz- + (_
n+1)/2qEaPbq -etc.
nin 3 2n .5
* I suppose no attentive person will find any difficultyin this. It is only saying that, supposing the
interval between nothing and certainty divided into a hundred equal chances, there will be 44 of them
for a less proportion of blanks to prizes than 9 to 1, 31 for a greater than 11 to 1, and 25 for some propor-
tion between 9 to 1 and 11 to 1; in which it is obvious that, though one of these suppositions must be true,
yet, having each of them more chances against them than for them, they are all separately unlikely.
t See Mr De Moivre's Doctrine of Chances, page 250.

THOMAS BAYES 315
By makinghere 1000 = p, 100 = q, 1100 =n, i I = z,
= 1Z
Mz=z = 1048808, Eavb,7 ih
Apq/ V(Kpq)'
h being the ratio whose hyperboliclogarithmis

1[111]
12 Lnp q
1 rl1 1 1 [1
360~
360 ln
~
13~1 -3q5
1260 Ln5 p5 q5
p3 q3
[1 1 11
and K the ratio of the quadrantal arc to radius; the formerof these expressionswill be foundto be
0 7953, and the latter0 9405 etc. The chance enquiredafter,therefore, is greaterthan 0 7953, and les
than 0 9405. That is; therewill be an odds forbeing rightin guessingthat the proportionof blanks to
prizesliesnearlybetween9 to 1 and 11to 1,(or exactlybetween9 to 1 and 1111to 99), whichis greaterthan
4 to 1, and less than 16 to 1.
Suppose,again, thatno moreis knownthan that blankshave been drawn 10,000timesandprizes 1000
timesin 11,000trials; what will the chance now mentionedbe?
Here the second as well as the firstrule becomes useless,the value of mz being so greatas to render
it scarcelypossibleto calculate directlythe series
mz -
m3z3
3 +
(n-2)
2n5
m5z5
2n. 5 -_etc.? J
The third rule, therefore,must be used; and the informationit gives us is, that the required
chance is greaterthan 0-97421,or more than an odds of 40 to 1.
By calculationssimilarto thesemay be determineduniversally,what expectationsare warrantedby
any experiments,accordingto the different numberoftimesin whichtheyhave succeededand failed;or
what should be thoughtofthe probabilitythat any particularcause in nature,withwhichwe have any
acquaintance, will or will not, in any singletrial,producean effectthat has been conjoinedwithit.
Most persons,probably,mightexpect that the chancesin the specimenI have givenwould have been
greaterthan I have foundthem. But this onlyshewshow liable we are to errorwhenwe judge on this
subject independentlyofcalculation. One thing,however,should be rememberedhere; and that is, the
narrowness
narrowness ~ interval
of the ~ ~~~I0 between -9 and 1 121, or bewe
between TI
10 + -I and 101 - 1 1 . Had
110 an Had this interval been
taken a littlelarger,therewould have been a considerabledifference in the resultsof the calculations.
Thus had it been taken double,or z = -5 , it would have been foundin thefourthinstancethat insteadof
odds againsttherewereodds forbeingrightin judgingthattheprobabilityofdrawinga blank in a single
triallies between1-?+ 1 and I -- 5-
The foregoingcalculationsfurthershew us the uses and defectsof the rules laid down in the essay.
'Tis evidentthat thetwo last rulesdo not giveus therequiredchanceswithinsuchnarrowlimitsas could
be wished. But hereagain it shouldbe considered,thattheselimitsbecomenarrowerand narroweras q is
taken largerin respectof p; and when p and q are equal, the exact solutionis givenin all cases by the
secondrule.These tworulestherefore afforda directionto ourjudgmentthatmay be ofconsiderableuse
till some personshall discovera betterapproximationto the value of the two series in the firstrule.*
But what mostofall recommendsthe solutionin thisEssay is,that it is compleatin thosecases where
information is mostwanted,and whereMrDe Moivre'ssolutionofthe inverseproblemcan give littleor
no direction;I mean, in all cases whereeitherp or q are ofno considerablemagnitude. In othercases, or
when bothp and q are veryconsiderable,it is not difficult to perceivethe truthof what has been here
demonstrated,or that thereis reasonto believein generalthat the chancesforthehappeningofan event
are to thechancesforitsfailurein thesame ratiowiththatofp to q. But we shall be greatlydeceivedifwe
judge in thismannerwhen eitherp or q are small. And tho' in such cases the Data are-notsufficient to
discoverthe exact probabilityofan event,yet it is veryagreeable to be able to findthe limitsbetween
whichit is reasonableto thinkit must lie, and also to be able to determinethe precisedegreeof assent
whichis due to any conclusionsor assertionsrelatingto them.
* Since thiswas writtenI have foundout a methodof considerablyimprovingthe approximationin
the second and third rules by demonstratingthat the expression 2Z/{1+ 2Eab + 2EaPbq/n}comes
almostas near to the truevalue wantedas thereis reason to desire,onlyalways somewhatless. It seems
necessaryto hintthishere; thoughthe proofof it cannot be given.


Barnard 1958

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Barnard 1958

Uploaded by

Copyright:

Available Formats

Biometrika Trust

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

STUDIES IN THE HISTORY OF PROBABILITY AND STATISTICS

THOMAS BAYES-A BIOGRAPHICAL NOTE

Bayes's paper,reproducedin the following pages,mustrankas one of the mostfamous

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

AN ESSAY TOWARDS SOLVING A PROBLEM IN THE

BY THE LATE REV. MR BAYES, F.R.S.

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

COR. 2. If therebe several independentevents,and the probabilityof each one be a, and

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

LEM. 1. The probability that the point o will fall

AC are so many inconsistent events; and this sum, m o.

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

xP+1 qx P+2 q(q-1) xV+3

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

XP+l1 qxP+2 q(q- 1) Xp+3 _q(q- 1) (q-2) xP+ + etc.

AfHO=p?1 p+2 2(p?+3)

q+1 q+2 2(q+3) 2.3(q+4)

q+1 (q+1)(q+2) (q+1)(q+2)(q+3)

xPrq = xP(1-x)2 = xP.{ I-qx +q (q 2) X2-etc.

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

and consequentlythe ratio of tFCfto ACFH is (n+ 1) E multipliedinto the difference

* ~~~~Xp+1qXp+2 q(q -1) Xp+3

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

my chance to be in the rightis greaterthan

and less than

and ifp = q my chance is 2X exactly.

1 [11 11 1 -11 1] 1 [1 111

c denotingthe circumference ofa circlewhoseradiusis unity. But MrBayes, in a precedingpaper in this

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

wherethe coefficients of B, C, D, E, F, etc. in the values of D, E, F, etc. are the 2, 3, 4, etc.

mz - im3z3 +(n -2)rm5Z5 etc.

2n 360n3+ 1260n5 1680n7+etc

(n+ 1)12 (n+ 2)2mz (n+ 2) (n+ 4) 4m3z3

(n + 2) (n+ 4) (n+ 6)8m5z5+ (n - 2) (n+ 4) (n+ 6) (n+ 8) 16m7z7

mz- m3z3 (n-2)" m5z5 etc.

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

V(Kpq) h J2H 12 (n + 1) (1- 2M2Z2/n)in+l1

2 <(Kpq) h {2H _2 (n + 1) (1 - 2m2z2/n)in+l 12 n (n + 1) (1 - 2M2z2/n)in+2

where in2,K, h and Hfstand forthe quantities already explained.

Containing an application of the foregoing Rules to some particular Cases

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

{XP+1 qxp+28 )12~~~~~~~~~~~~~~~~~~~~~~~x+1

He will,therefore, have a betterchance forbeingrightthan in the formerinstance,the odds against

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

h being the ratio whose hyperboliclogarithmis

This content downloaded on Wed, 2 Jan 2013 16:27:51 PM

You might also like