Early History of the Theory of Probability

O. B. SHEYNIN

Communicated by A . P. YOUSCHKEVITCH

Contents
1. I r t m d u c t i o n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
2. O r i g i n o f Stochastic I d e a s a n d N o t i o n s in Science a n d Society . . . . . . . . . . . . . 202
2.1. G a m e s o f c h a n c e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
2.2. J u r i s p r u d e n c e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
2.3. I n s u r a n c e o f p r o p e r t y a n d life . . . . . . . . . . . . . . . . . . . . . . . . . 206
2.3.1. I n s u r a n c e o f p r o p e r t y . . . . . . . . . . . . . . . . . . . . . . . . . . 206
2.3.2. Life i r s u r a ~ c e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
2.3.3. J. DE WITT'S m e m o r a n d u m . . . . . . . . . . . . . . . . . . . . . . . . 212
2.4. Political a r i t h m e t i c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
2.4.1 A n c i e n t a n d a n t i q u e h i s t o r y . . . . . . . . . . . . . . . . . . . . . . . 216
2.4.2. PETTY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
2.4.3. GRAUNT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
2.4.4. I_EI~ NIZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
2.4.5. HAl L~Y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
2.4.6. N E t M~ NN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
3. PASCAL a n d FE~M~T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
3.1. FERMAT-PASCAL, letter w i t h o u t d a t e . . . . . . . . . . . . . . . . . . . . . . . 232
3.2. FASCAL-FERMAT, letter d a t e d 29 July, 1654 . . . . . . . . . . . . . . . . . . . . 232
3.2.1. R a n d o m w a l k s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
3.2.2. B i n o m i a l ceefficients . . . . . . . . . . . . . . . . . . . . . . . . . . 234
3.2.3. Small differences b e t w e e n probabilities . . . . . . . . . . . . . . . . . . 235
3.3. FASCAL-FERMAT, letter d a t e d 24 A u g u s t , 1654 . . . . . . . . . . . . . . . . . . 236
3.4. FERMAT-PASCAI, k t t e r d a t e d 25 S e p t e m b e r , 1654 . . . . . . . . . . . . . . . . . 237
3.5. A n o t h e r p r o b l e m p o s e d by PASCAL . . . . . . . . . . . . . . . . . . . . . . . 238
3.6. PASCAL'S treatise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
3.7. A l e a e g e c m e t r i a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
4. H u ~ GE~,S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
4.1. HUYGtNS' maSn w o r k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
4.2. HIJYGENS' c o r r e s p o n d e n c e a n d m a n u s c r i p t s . . . . . . . . . . . . . . . . . . . 241
4.2.1. T h e 3 e a r 1656 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
4.2.2. Tlne } e a r 1665 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.2.3. T h e } e a r 1669 ( m o r t a l i t y ) . . . . . . . . . . . . . . . . . . . . . . . . 247
4.2.4. W o l k d u r i n g 1 6 7 6 - 1 6 8 8 . . . . . . . . . . . . . . . . . . . . . . . . . 249
4.2.5. HU'~GE>S' a n a l y t i c a l m e l h o d s . . . . . . . . . . . . . . . . . . . . . . 250
4.3. M o r n c e r t a i n t y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
5. G e r e r a l C o n c l u s i c n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
A d d e n d u m to ~ 2.2, 2.3.2, 2.3.3 a n d 2.4.4 . . . . . . . . . . . . . . . . . . . . . . . 254
Referen(es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
202 O.B. SrIEYNIN

1. Introduction
The theory of probability originated in the period f r o m 1654 (correspondence
between PASCAL and FERMAT on probability) to 1713 (posthumous publication
of J. BERNOULH'S Ars conjectandi [3]). The present paper is devoted to this period,
i.e. to the work of PASCAL and FERMAT (§3) and HUYGENS (§4) 1. However, in
order to assess the share of various factors in the emergence of the theory, a section
(§2) is included on the relevant history of games of chance, jurisprudence, in-
surance of life and property as well as political arithmetic and demographic
statistics. Actually, §2 continues my own previous article [93]. Lastly, general
conclusions are formulated in § 5.
I do not deal with lotteries; to my mind, the study of expected gains presented
no difficulty and hence made no contribution to probability 2. I shall only mention
the famous Genoese lottery devised in imitation of the annual city elections, in
which five candidates had to be selected from a hundred. The electoral procedure
consisted in casting lots and from the beginning of the 17 th century at the latest
bets were made on their results. Local bankers offered twenty thousand-fold
payoff to anyone guessing all five successful candidates ([10], p. 57) 3. Around
1620 the c o m m o n lotto originated from this lottery ([11], p. 336). Stochastic
calculations related to this lottery are due to N. BERNOULLI [4], [7]; see also §2.2.
I do not mention NEWTON. NEWTON'S philosophy did have an extensive,
though indirect influence on probability. Actually NEWTON'S ideas rather than
games of chance per se turned DE MOIVRE's attention to probability and directed
him in his stochastic research ([89], pp. 230-231). In addition, NEWTON deserves
credit for a number of achievements in probability and stochastic considerations
in astronomy (ibidem, pp. 217-227). NEWTON'S work, however, was not carried
further in the period under review.
As I see it, the main importance of this article lies in the detailed description
of its subject with some new findings, or conjectures framed, in §§ 2.4.3, 2.4.4,
3.2.3 and 4.2.3.
2. Origin of Stochastic Ideas and Notions in Science and Society
2.1. Games o f chance
As I have noted ([93], p. 114), games of chance p r o m o t e d the general, possibly
intuitive idea of definite stochastic properties evinced by mean outcomes, and
(10. 113) served to prove that certain events in nature were designed rather than
produced by chance.
In games of chance PASCAL, FERMAT and HUYGENS were confronted with
problems whose solutions gave rise to stochastic theory. Moreover, in their
efforts to assess the potentialities of the emerging theory as well as their own
competence, scholars directed their attention to various problems encountered
in games of chance; see for example §4.2.2. Lastly, games of chance supply part
1 The same period witnessed the publication of DE MOIVRE'Sfirst research in probability [20].
On the whole, however, his works date later. The same applies to a few articles by J. BERNOULLI.
2 Calculations of the probabilities of sequences in extracted numbers; of extracting all the num-
bers of a lottery in a given number of drawings at least once etc. are of course sufficientlyinteresting.
This subject was not investigated until later, approximately from the middle of the 18th century.
3 A good example of downright robbery: see § 2.3.2.
 Early History of Probability 203

of the subject of HUYGENS' treatise [48] and the entire subject of a book by
MONTMORT [691.1
LEIBNIZ ([93], p. 115) even supposed that games of chance could be used as
models for studying the Erfindungskunst and, I would add, for originating a
statistical decision theory.
All this is not difficult to explain: games of chance, and possibly only they,
could at that time provide models for posing natural and properly formulated
stochastic problems. Besides this, studies of games of chance were in the social
order of the day.
The possibility of other applications for probability was contemplated by
HUYOENS (§4.1). His opinion is not convincing, though, since there is not a
hint of other applications in his treatise [48]. But then, games of chance do not
even occur in the work of DE WITT (§2.3.3) and HALLEY(§2.4.5).
However, bearing in mind that HUYGENS' treatise represented merely the
infantile stage of probability and that J. BERNOULLI'SArs conjectandi was con-
ceived as a treatise on probability with applications aufgesellschaftliche und wirt-
schaftliche Fragen (author's letter to LEmNIZ, 3 Oct. 1703, [33], p. 404), we may
say that the emergence of probability was very little indebted to games of chance2.
It is not to be inferred though that after J. BERNOULLIgames of chance faded
into insignificance. On the contrary, the study of these games proved rather
fruitful for DE MOIVRE. True though after DE MOIVREthe role of games of chance
diminished, yet even now they enjoy certain methodological importance in
probability proper to say nothing about the theory of games.
Considering the early history of games of chance KENDALL ([54], p. 26)
remarks that
by the end of the fifteenth century the foundations of a doctrine of chance was
being laid. The necessal T conceptualization of the perfect die and the equal frequency
of occurrence of each face are explicit.
Why then, asks K E N D A L L , did not the theory of probability emerge in those times ?
Listing several possible reasons (p. 30)3 he concludes that
It is in basic attitudes towards the phenomenal world, in religious and moral
teachings and barriers, that I incline to seek for an explanation of the delay.
Another explanation is by MAISTROV ([66], chap. 1, §2) according to whom
the only essential reason here was the absence of any relevant practical demand.
1 Especially interesting in this book is the game called le her. Understandably, both MOtqTMORX
(p. 278) and N. BERNOULLI (ibidem, p. 334) failed to investigate this game properly; such a study
would have required the use of the minimax method ([32], p. 158).
2 In itself, this assertion is not original ([66], chap. 1, § 2).
3 KENDALL connects one of these reasons (superstition) with the psychology of gamblers. How-
ever, psychology may also enter in a different way. Recent experiments [14], [15] seem to suggest
that psychological subjective probabilities differ from objective statistical probabilities. In one ex-
periment the subjects were asked to participate in one of two lotteries choosing either one ticket of
lottery A the probability of winning in which was 0.1 or several tickets of lottery B with respective
probability 0.01. Asked to say how m a n y tickets of lottery B they were prepared to have instead of
one ticket of lottery A, m a n y subjects chose a n u m b e r considerably different from ten. On superstitions
see MONTMORT([69], pp. vi--viii) and LAPLACE(Essaiphilosophique, chap. Des illusions dans l'estimation
des probabilitds).
204 O . B . SHEYNIN

Nevertheless, it is possible to imagine the beginnings of a probability calculus

(say, the appearance of a treatise resembling that of HUYGENS) emerging in the
middle of the 16th century. Why did nothing of the kind happen? In my opinion,
because not the 16TM,but the 17th century marked the beginning of modernity both
in society and science. During the 17th century scientific communities became
influential and scientific correspondence essentially expanded; both these factors
taken together actually midwifed the development of probability (see §§ 3-4).
As to religious and moral obstacles in the path of probability, it would be
more proper to speak about hindrances to a philosophical apprehension of such
concepts as randomness and probability. These obstacles were occasioned [93]
by the general state of philosophy which, even in the 17t~ century, did not com-
pletely abandon the obsolete ARISTOT~LIANpicture of the world.
A venerable problem in games of chance formulated in 1380 or even before
([73], p. 414) was the problem of dividing the stakes (or, as it is usually called, the
problem of points). Two gamblers start playing. The game is to continue until
one of them scores n points; however, for some reason it is interrupted on score
a : b (a, b < n). The problem of dividing the stakes in such cases 1 had been con-
sidered by C A R D A N O , TARTAGLIAand PEVERONE. In 1558 PEVERONE ([54], p. 27) 2
gave a correct answer for problem (10; 8:9) and, were it not for his blunder, he
would have given a correct answer for problem (10; 7:9). The further history of
the problem of points is continued in § 3.

2.2 Jurisprudence.

It is my understanding that, beginning approximately from the second half

of the 17th century, the importance of civil suits considerably increased and the
practice and, possibly, theory of legal proceedings, both criminal and civil, more
and more openly employed stochastic estimates of proof thus disseminating
stochastic ideas and notions.
At any rate, it is generally known that LEIBNIZ ([93], p. 109) testified to the
existence of an elementary scale of stochastic proofs in jurisprudence.
DESCARTES ([23], p. 323) introduced the
cOrtitude morale, suffisante pour regler nos mceurs, ou aussi grande que celle des
choses dont nous n'avons point coustume de douter touchant la conduite de la vie,
bien que nous sachions qu'il se peut faire, absolument parlant, qu'elles soient
fausses 3.

It is possible that D~SCARTES held this cOrtitude morale to be used in juris-

prudence. A similar opinion was held much earlier by THOMAS A Q U I N A S ([93],
p. 108) :
In the business affairs o f men we must be content with a certain conjectural
probability.

1 Hereafter, I shall use notation (n; a : b ) to describe the initial conditions of the probiem.
2 I have not found PEVERONE either in POGGENDORFF or in some of the general Italian encyclop-
paedias. According to H.O. LANCASTER, Bibliography o f statistical bibliographies. Edinburgh a.o.,
1968, p. 22, an article devoted to G.F. PEVERONE (1509-1559) is in A t t i Torin Accad. 17, 1882, 320-324.
3 L'autre sort de certitude, continues DESCARTES (p. 324), est lors que nous pensons qu'il n'est
aucunement possible que la chose soit autre que nous la jugeons.., cette certitude s'estend g~ tout ce qui
est demonstrO dans la M a t h e m a t i q u e . . .
 Early History of Probability 205

Reasonings on moral certitude are also contained in part 4 of the Logique de

Port-Royal [2]. In particular, chap. 15 cites an instance of applying this certitude
in a legal case 1. This example was later borrowed by J. BERNOULLI ([3], pt. 4,
chap. 3).
LEIBNIZ, for his part, thought of applying moral certitude in theology. The
extant contents of one of his unwritten works [57] mentions a chapter Demon-
stratio probabilitatis infinitae, seu certitudinis moralis, qubd pulcritudo mundi
oriatur ?t mente. I do not know whether LEmyIz proposed to use moral certitude
in natural science but at least he stated ([111], p. 169, this being a translation
from Latin, from un coupon ... porte au verso un brouillon de lettre date ... i693
([/8], p. 232)) that there exist Drei Grade der Sicherheit in Urteilen: die logische
Gewiflheit, die physische Gewifiheit--die nur eine logische Wahrscheinlichkeit
ist--, die physische Wahrscheinlichkeit. Ein Beispiel... der dritten." "Der Siidwind
ist regnerisch", welche [Siitze] meistens wahr sind, obwohl sic nicht selten fehl-
gehen. See §4.3 for a discussion of the relevant opinion of HUYGENS. LEIBNIZalso
held ([58], p. 453) that scientia est notitia certa ... Opinio est notitia probabilis ...
And, on p. 457: Certitudo est claritas veritatis.
Applications of probability to sujets politiques, ceconomiques ou moraux were
also contemplated by MONTMORT ([69], p. xiii).
Ce qui m'en ernp~chO, he complains, c'est l'embarras oft je me suis trouvO de
faire des hypotheses, qui ~tant appuyOes sur des faits certains ...
An example of such a hypothesis is the one adopted by D. BERNOULLI(in 1738)
in connection with his moral expectation.
During 1654-1713 only one work, a dissertation by N. BERNOULH [4], was
written on the application of probability in jurisprudence. In this work BERNOULLI
(1) Calculates, from GRAUNT'S mortality table (§2.4.3), the mean duration of life
for various ages.
(2) Recommends the use of mean duration of life for computing the value of an-
nuities and estimating the probability of death of absentees about whose existence
nothing is known 2.
1 Mathematicalexpectationisintroducedinchap. 16, p a r t 4 o f L o g i q u e w h i c h p o s s i b l y f u r t h e r e d
the dissemination of stochastic ideas. Declaring that this concept should govern everyday behaviour
(moral certainty is no longer mentioned), the authors agree with the logic of PASCAL'S wager (with
his demonstration of the benefit accrued by believing in God). This wager ([32], p. 154; [43], pp.
63-72) was published posthumously in PASCAL'S PensOes.
Such absentees, claims B~RNOULLI, should not be declared dead until the probability of their
death does not become twice as large as the probability of their being alive. This proposal was some-
what developed by CONDORCET [105]. Regarding the absentee's property, CONDORCET considered
the ratio of the risks of its being lost either by the absentee himself or by his heirs. An explicit or
implicit comparison of two probabilities, or risks, was always inherent of jurisprudence.
Absentees are mentioned in Roman law ([67], p. 145) :
Une loi de Julien ddcida que la captivitd ne serait plus une cause de dissolution du mariage et que
l'incertitude m~me de l'existence de l'@oux prisonnier ne pourrait permettre d son conjoint de se remarier
qu'aprOs cinq ans d partir du jour de la captivitd.

It seems however that legislation of this kind was always based exclusively on common sense. Thus,
even in the 19th century (ibidem), relevant stochastic considerations were extremely simple:
L e e [ode] cir. a divisO l'absence en trois p~riodes: 1 ° La prOsomption d'absence. Pendant cette
pOriode le doute sur l'existance de l'absent est trOs ldger...2 ° La d~elaration d'absence. Pendant cette
p~riode la prdsomption de mort l'emporte sur la prOsomption d'existance . . . . 3 ° L'envoi en possession
d~finitif. Avec le temps la pr~somption de mort se fortifie et se change presque en eOrtitude.
206 O.B. SHEYNIN

(3) Calculates the mathematical expectation of losses in marine insurance.

(4) Calculates the mathematical expectation of gains in the Genoese lottery [7].
(5) Calculates the probability of truth of testimonies.

I must draw particular attention of the fact that BERNOULLI determined life
expectancy of the last survivor of a group of men ([96], §340). Assuming a con-
tinuous uniform distribution, he calculated the expectation of the corresponding
order statistic. Thus he was the first to use, in a published work, both the former
and the latter 1.
It is doubtful whether BERNOULLI'Sdissertation had influence on jurisprudence
but it certainly became known to such scholars as CONDORCET, LAPLACE and
PoISSON whose work brought about a general advancement in probability if not
in jurisprudence 2.

2.3. Insurance o f property and life

2.3.1. Insurance of property. Insurance of property has existed since ancient
times. Thus ([84], p. 40)
two thousand years B. C . . . . participants o f trade caravans in the Near East
concluded agreements to share damages incurred en route due to robbery, theft or
loss. And, according to the Talmud, similar agreements were concluded in Palestine
and Syria ...
Similar agreements, continues the author, were also concluded between merchants
active in marine commerce on the shores of the Persian Gulf, in Phoenicia and
ancient Greece while the Solon law mentions companies a m o n g the participants
of which existed agreements to share damages incurred in marine commerce
(and sea robbery, the author adds).
The agreements mentioned lack stochastic ideas and notions; they do not
even embrace any system of initially determined insurance payments. It seems
that such a system came into being in European feudal guilds; also, in Japan a
similar system of insurance payments existed even in the 12 th century [71].
Marine insurance should be dealt with separately. Possibly until the 19 th
century ([13], p. 349)
L'assurance maritime a ~t~ et devait ~tre la premiere f o r m e de l'assurance.
A certaine organisation for such insurance came into being in the 14th century,
while during the 15 th century (ibidem, p. 350)
l'assurance (maritime)fait ddj& l'objet de dispositions l@islatives importantes.
However, the forms of marine insurance were not conducive to stochastic reason-
ings. One of these forms was ([25], p. 4)

i The relevant work of DEWITT(§ 2.3.3) and HUYGENS(§ 4.2.3) remained unpublished.
2 CONDORCET([17], p. 498) even overestimated N. BERNOULLI'Swork; without expresslymention-
ing it he said: Depuis l'ouvrage de N. Bernoulli, le calcul desprobabilitOs est devenu l'objet des recherches
des philosophes eomme des travaux des math~maticiens... Such an opinion should have been reserved
for J. BERNOULLI!
 Early History of Probability 207

l'assurance par forme de gageure. And ([13], p. 349) On rencontre ... dans les
textes du Digeste la stipulation suivante." "Je stipule que vous me donnerez 100,
si tel navire n'arrive pas d ' A s i e " . . .
... il seroit odieux ([25], p. 6) qu'on se m~t clans le cas de desirer la perte d'un
vaisseau ... clans la plfipart des Places de Commerce, les Assurances par gageure
ont OtO prohib~es.
Nevertheless ([13], p. 349) this form of insurance air reparu ~ diffOrentes Opoques
avec unefdcheuse persistance and, in the 16th century (ibidem, p. 351), L'assurance
(maritime) d~gOndra rite en opOration de j e u du caractOre le plus al~atoire ...
Another form of marine insurance was so-called bottomry--perfectly legal but
also primitive. Bottomry ~ implied a mortgage on a vessel with the stipulation that
the repayment of loan is conditional on the safe arrival of the vessel. The interest
on bottomry was considerably higher than on loans in general ([84], pp. 68-7.0;
[45], pp. 127-131).
On the brighter side, I shall quote a particular statement and a noteworthy
description of what possibly was the birth of the modern form of marine in-
surance. The statement, regrettably unsubstantiated, is to the effect that marine
insurance almost gave rise to the notion of probability (or expectation?) of a
random event ([13], p. 349):
Au moyen ~ge, pour la premiOre fois, on a compris que la risque est une rOalitO
que l'on peut s~parer idOalement du tout dont elle fait partie pour lui assigner sa
valeur propre ...
The description ([45], pp. 141-142) refers to the speech of the Lord Keeper
BACON (father of Sir FRANCIS)in 1558:
Doth not the wise merchant in every adventure o f danger, give part to have the
rest assured?
The same author also quotes from the first English Statute on assurance:
A n d whereas it hathe bene tyme out o f mynde an usage amongste merchantes,
both o f this realme and o f forraine nacyons, when they m a k e any great adventure ...
to give some consideracion o f money to other persons ... to have f r o m them assurance
made o f their goodes, merchandizes, ships, and things adventured, ... whiche course
o f dealinge is commonly termed a policie o f assurance ...
This is a passage from Publicke Acte No. 12 (1601)--An Acte conc'ninge matters
o f Assurances amongste Marchantes 2. The main text of the Act is devoted to the
legal aspect of disputes on policies of assurance.

2.3.2. Life insurance. A policy on human life (is) defined as any instrument by
which the p a y m e n t o f money is assured on death or the happening o f any contingency
dependent on human life, or any instrument evidencing a contract that is subject to
p a y m e n t o f premiums f o r a term dependent on human life 3.

1 Enc. Brit., vol. 4, 1965,p. 8. The author continues: a similar contract creating a security interest
in cargo is (was?) called respondentia.
2 Statutes of the realm, vol. 4, pt. 2, pp. 978-979.
Ene. Brit., vol. 13, 1965. Life insurance (p. 1091).
208 O.B. SHEYNIN

The second form of life insurance is a life annuity offered either individually or
to a group of men. F o r example, mutual insurance in tontine associations 1 secured
an annuity changing (increasing) with time, or, more precisely, with the decrease
of the number of insured still alive. Also forms of mutual insurance (usually of
m a n and wife) existed with a constant annuity payable until the death of the last
survivor 2. The same author continues (p. 1094):
The p a y m e n t o f certain benefits on death against certain periodical subscriptions
are to be f o u n d in the R o m a n collegia (artisans' associations).
This statement does not directly contradict the opinion of another author
([13], p. 348):
La preuve absolue que les Romains n'ont pas connu l'assurance, c'est qu'on ne
trouve pas un seul m o t relatif d ce contrat dans les ~crits de leurs jurisconsultes ...
In any case, the system of periodical subscriptions seems to have faded out
of existence in later centuries. Thus, for example ([84], p. 61) in the 13 th century
Danish guilds provided insurance based on subsequent distributions of damages
(e.g., from shipwrecks or captivity with ensuing p a y m e n t of ransom). Allowance
for ransom m o n e y is actually an insurance premium.
During the Middle Ages (ibidem) mutual guild insurance covered most
diversified cases, including those directly related to the personality of guild
members. A definite example follows (p. 62): in 1284 one of the English guilds
paid allowances in cases of incurable diseases or blindness.
Speaking about insurance against accidents and against sickness, another
author ([41], p. 74), without specifying the system of subscriptions, says:

... whereas we [in England] began to be busy in this direction about the middle o f
the sixteenth century (1560) Italy practised this civilising art o f insurance as early
as the end o f the twelfth century ...
Obviously, life insurance had gradually spread beyond the limits of separate
guilds. 3 Thus, an a n o n y m o u s French author of the 16 th century testifies ([45],
p. 228) :
Pilgrims going to the H o l y Sepulchre o f Jerusalem, or on other distant voyages,
m a y effect insurance f o r their redemption . . . . A n o t h e r k i n d o f insurance is made by
other nations upon the life o f men, in case o f their decease upon their voyage ...
Which are all stipulations forbidden ...

A name derived from that of the inventor, the Italian LAURENSTONTI[46].

2 If primitive forms of early lifeinsurance are disregarded, a condition possibly not even necessary,
life annuities would constitute chronologically the first form of life insurance and this is how I shall
call them. Insurance against death in my terminology would be the second form of life insurance.
3 A similar development took place in Japan. Considering insurance both of life and property,
NOGUCHI([71], p. 238) maintains:
Wie in Europa im Mittelalter die Gilden Trdger der Versicherung waren, so hat sich in Japan schon
fiber tausend Jahre friiher derselbe Gedanke und dieselbe Organisation gegenseitiger Hilfe durchge-
setzt.
And, on p. 242, regrettably with no reference added:
Nun sind die meisten Forscher in Europa der Meinung, das Versicherungswesenhabe sich, was die
Entstehung des Gedankens betrifJL aus dem Gildenwesen entwickelt, wdhrend die Entstehung der Form
auf die Seeversieherung zuriickgehe. Genau das gleiche ist aueh in Japan der Fall.
 Early History of Probability 209

It seems that the main reason for forbidding insurance of life was its developing
connection with gambling 1. Thus, quoting numerous references, I~MERIGON([25],
p. 198) says:
... ces sortes d'Assurances ne sont pas des Assurances proprement dites ," ce sont
de vOritables gageures ... Ces gageures ... sont prohibdes en Hollande, & en plusieurs
autres pays ... Depuis longtemps elles avoient prohibOes en France ...
The Amsterdam Ordinance of 1598 ([45], p. 229) expressly (prohibited)
insurance o f life of any person and likewise wagers upon any voyage ... Similar
prohibitions (ibidem) were contained in the Rotterdam ordinances of 1604 and
1635, in the Marine ordinance of Louis XIV (1681) and in a series of Netherlands'
ordinances issued in 1570-1635, while in the Statutes of Genoa for 1588 ([54],
p. 32) insurance of life was forbidden sine licentia Senatus ... Some of the passages
quoted above suggest that insurance of life could well have originated from the
semilegal and odious marine insurance parJ'orme de gageure (§2.3.1) 2.
Possibly numerous prohibitions hindered the advancement of life insurance
in the second form. In any case, though, it seems that no legal prohibition ever
applied to life insurance in the form of annuities (tontines excluded) 3 which
existed even in ancient Rome ([45], p. 224):
... the Roman lawyers, at least about the time o f the division o f the Empire,
found it necessary to consider and frame a table by which annuities could be valued
so as to meet the requirements of the Falcidian law, which prevented the testator
from leaving more than three-quarters of his property to any others than legally
constituted heirs ... one o f the most eminent commentators on the Justinian Code,
the Praetorian Praefect Ulpianus (170-228), gave a table o f the estimated present
worth of... life annuities.
This table was based on a table of expectations of life (see my Table 2.3.2 (1). In
the opinion of HENDRIKS (pp. 224--225) ULPIANUS may have obtained these
expectancies either

Table 2.3.2 (1)

Expectation of Life of Males and Females According to ULPIANUS ([45], p. 225)

Age Expectation Age Expectation Age Expectation

0-20 30 41-42 18 47-48 12

20-25 28 42-43 17 48-49 11
25-30 25 43-44 16 49-50 10
30-35 22 44-45 15 50-55 9
35-40 20 45-46 14 55-60 7
40-41 19 46-47 13 60-... 5

1 See also § 4.2.3 where I quote a relevant remark of HUYGENS.

2 Wide research concedes that Life Insurance came into its own, not by a front-door entrance,
but by the marine insurance porthole ([72], p. 78). The text of O'DONNELL'S a b u n d a n t and fascinating
non-mathematical source, which I did not explore, carries no references to the appended list of
literature.
3 More precisely ([114], pp. 484-485), Eine pdpstliche Bulle yon 1423 erkldrt schliesslieh (after
about a century of prohibition) den Rentenkauffiir erlaubt.
210 O.B. SHEYNIN

f r o m inquiries on the results o f like annuity engagements, or f r o m returns o f the

number o f deaths occurring within a given time at various ages; the f o r m e r m e t h o d
would seem to have been the most likely to be available, but the other was quite
within the bounds o f possibility as the foundation o f an approximate computation,
f o r there is ample record o f a kind o f registration, or ephemeris, o f deaths having
been observed by the ancients.
On the contrary, GREENWOOD ([38]; [39], p. 67) asserts that ULPIANUS did
not base his table on any statistical data whatsoever. But even in this case (and
even bearing in mind that ULPIANUS' expectation of life does not necessarily
coincide with its expectation as understood nowadays) ULVIANUS' table at least
methodologically constituted the highest achievement of demographic statistics
until the 17 th century (see also §2.4.1). Regrettably, owing to general conditions
of the development of society and science, this table was forgotten.
In modern times annuities are known at least from the 14 th century 1. Referring
to a series of Dutch sources published in 1670-1671, HENDRIKS ([45], p. 112)
arrives at a conclusion that in Holland, both long before, and at that time, the
price of annuities normally did not depend on the age of the annuitant. This
practice, as HACKING ([43], p. 112) seems to imply, was borne out by high mortality
both in childhood and in old age. That annuities were actually purchased in the
name of children and even infants is testified by HUDDE ([47], t. 7, pp. 95--96) 2
who compiled statistics showing the age of annuitants
sur les t~tes desquelles des contrats de rentes viag&es ont ~t~ vendus par le
gouvernement des Provinces unies, en 1586, 1587, 1588, 1589 et 1590.
A very numerous group constituted annuitants aged from two to seven years.
HUODE also shows the duration of life of each annuitant so that his is a longevity
table.
On the other hand, prices of annuities sold in Holland in 1672 and 1673 did
depend on the age of annuitants ([16], p. 1205). It is another question whether
and to what extent the need to calculate the price of annuities as against age of
annuitants stimulated statistical research of mortality (see also §2.3.3).
As to tontines, it seems that they were neither socially accepted nor widespread
([45], p. 116):
... in England we have such strong prejudices against Tontine Associations,-
based, perhaps, on the assumed rationale that they are too selfish and speculative to
be encouraged...
The same feeling prevailed in France ([30], p. 617):
L e Ministre de l'Int~rieur ... a d&irO que l'Acad~mie des Sciences choisit ... une
Commission charg& d'examiner les articles qui rOglent les int(rOts respectifs des
aetionnaires (of a proposed tontine).

i ...der 61teste reine Leibrentenvertrag wurde im Jahre 1308...abgesehlossen ([114], p. 484); in

another source ([113], pp. 186-193) Certificats de rente viag&e dating back to 1228 and 1229 are
published. One exampleof an annuity on life regards BENVENUTOCELLINI([12], p. 423; [1]): I sacrificed
my pay for his portrait ... and he arranged that he should keep my money at 15 per cent during my
natural life.
z HUDDEwill be mentioned time and time again. There exists a general description [42] of his
mathematical works.
 Early History of Probability 211

Des tontines, continues this author (FOURIER) o n p. 619, exercent deuxpenchants

funestes : l'un est la disposition gl attendre du hasard ce qui devrait ~tre le fruit d'une
industrie profitable gl tous, ou le rdsultat ordinaire des institutions; l'autre est le
ddsir d'augmenter ses jouissances personelles en s'isolant du reste de la societal.
He brings forward a supplementary consideration (pp. 625-626): the Academy,
having been consulted
par le Gouvernement sur Ie projet de l'3tablissement de la Caisse dite de Lafarge 1,
proposa un avis contraire gt ce projet. Nous avons trouvO dans nos Archives le Rap-
port de la Commission charg~e de l'examen de cette question; il a ~t~ adoptO (en)
1790: il est sign~ de M M . de Laplace, rapporteur, Vandermonde, Coulomb, Lagrange
et Condorcet.
The final conclusion is obvious (p. 630) :
... l'Acaddmie ne peut que refuser son approbation gl un ~tablissement irr@ulier,
contraire aux rues du Gouvernement, et m~me aux intentions des auteurs du projet
(because, see p. 629, le placement des capitaux en tontine est beaucoup moins
favorable que le simple contrat de rente viagOre).
L'AcadOmie approuve le Rapport et en adopte les conclusions.
Nevertheless, tontines did exist in the 17th century; thus in France, tontines
were established in 1689 and 1696 ([46], pp. 206 and 209; [95], p. 405) and in
Holland, in 1671 ([94], p. 226). The down payment in French tontines depended
on the age of annuitants ([95], p. 405; [4], p. 35) necessitating that the law of
population mortality be taken into consideration 2. In their turn, tontines provided
statistical data on mortality; HEZqDRII~S([45], p. 116) even supposed that this data
was used for compiling first tables of mortality. So it possibly was, with the notable
exceptions of such tables as GRAUNT'S (§2.4.3) and HALLEY'S(§2.4.4).
Societies which provided life insurance in the second form came into being
in the 18th century ([10], p. 100):
Zwar kann eine 1699 in England unter dem Namen der Society o f Assurance f o r
widows and orphans . .. ins Leben gerufene Gesellschaft nicht als Lebensversicherungs-
anstalt im heutigen Sinne bezeichnet werden.
(Continuing his account, the author substantiates his opinion.)
Another author 3 states that the first to be established (in 1706) 4 was the
Amicable society. He also notes that before the 18 th century insurance (in the
second form) was practised mostly so as to cover contingencies of a temporary
character and even these only over short periods.
1 j . Lafarge put ~tablir...une caisse d'~pargne, la c~l~bre ~tontine Lafarge~, dont le succ~sfut
dphOmOre, mais qui servit de moddle aux premiOres caisses d'Opargne... (Grand Larousse enc., t. 7,
1962, p. 542).
2 Data pertaining to early tontines were used by SEAL [88] in order to study laws of mortality.
SEAL maintains that Dutch annuity schemes of the 16th century were really tontines. HUDDE'S letter
to which both SEAL and myself refer is written in vernacular and I am unable to check this assertion.
Drawing on DEPARCIEUX, SEaL also uses data on mortality of French m o n k s during 1607-1669.
3 Enc. Brit., vol. 13, 1881, p. 180.
4 In the opinion of CHAUFTON([13], p. 351) the first successful life insurance society was established,
in England, in 1720.
212 O . B . SHEYNIN

The premiums were very high, adds he, but this was in part necessary for two
reasons." first, the insurers had no sufficient data upon which to estimate the risk
they incurred; and secondly, the transactions were probably not numerous enough
to secure anything like a regular average in the occurrence of claims.
These two reasons seem to allow, at least partly, for the fact that life insurance in
its second form did not play any important role in promoting stochastic ideas and
notions. The same conclusion, though for another reason, is shared by MRO6WK
([70], p. 50):
... neither joint-stock societies, nor banks, nor stock exchanges stood in need of
probability; their demands on probability appeared only in the 19 th century, when
methods of downright robbery had been superseded by those o f scientific gain.
There is ample evidence of cheating, if not of robbery, in life insurance both in
the 18th and in the 19th centuries 1. However, the first two reasons seem to be more
important; actually, they amount to saying that, until the 19th century, life in-
surance in its second form just was not sufficiently developed 2.
2.3.3. J. De Witt's memorandum [100]. A special feature c o m m o n to both
forms of life insurance should be mentioned: statistical data, and methods of
actuarial calculations could well have constituted a commercial, or even state
secret. At least this is evident as regards the m e m o r a n d u m of DE WITT [100], a
prominent mathematician and statesman ;3 see below.
In this memorandum, adressed to the members of the governement of the
Provinces unies4, DE WITT strove to substantiate the possibility of raising the price
of life annuities sold by the state. Assuming definite suppositions regarding the
law of mortality and, also, a 4 ~ per annum (or, rather, a (1/1.04-1) 100~ per
half year) discount, DE WITT, in his own words ([45], p. 234),
mathematically ... proved that ... the life annuity should be sold at 16 years'purch-
ase.
1 A number o f minor companies became defunct during the period ofspeculativefinancial schemes
which resulted in the crisis o f 1720...Speculation was common throughout the century...between 1800
and I870 some 500 new offices were established...some were of definitely fraudulent intent. (Eric. Brit.,
vol. 13, 1965, p. 1094).
2 Without discussing the history of insurance BIENAYMt6 [6] remarked that c o m p o u n d interest
adversely influences the activities of insurance societies: owing to c o m p o u n d interest even a small
loss incurred during the initial period of work of such societies would not be compensated by a later
gain of the same order. BIENAYN£ concludes that success in insurance is only possible when trans-
actions are sufficiently numerous.
3 Utterances of various scholars about DE WITT were collected by HENDRIKS ([45], pp. 253--255).
To these utterances I shall add the one contained in the correspondence of HUYGENS ([47], t. 2, pp.
411--412) dating to 1659: II est bien scavant en la Geometrie et en l'algebre et s'y exerce tousjours
non obstant les grandes affaires qu'il a sur les bras.
There is also a m o d e r n biography of DE WITT [87].
4 Or at least to Noble and mighty Lords of the state. See p. 232 of an English translation of DE WITT
due to HENDRIKS inserted into his contribution [45]. HENDRIKS, as he says on his p. 257, had discovered
DE WITT'S m e m o r a n d u m in Resolutions o f the States o f Holland and West Friesland o f 1671 (he mentions
this English title only). Most likely, the Resolutions contain a reprinted version of the m e m o r a n d u m ;
in any case, HENDRIKS (p. 257) mentions a misprint in calculations which does not occur in the version
[100] I myself saw.
 Early History of Probability 213

Table 2.3.3 (1)

DE WITT'STable ([100], pp. 16 20).Calculation of Price of Life Annuities. NominalAnnuity 20.106
Stuyvers
Chances Of stuyvers Half years Chances Of stuyvers Half years
1 9,805, 807 1 ~ 455, 999,472 119
1 19,421, 192 2 ½ 456, 950, 076 120

1 . . . . . . . . . . . . . . . . . . . .4. .3. 2. .,. . .4.9. .0. ,. . 8 2 5 ;; ........ .......... i; i;080 ........... ......

433,897,951 100 ½ 472, 523, 275 140

1
3 479, 820, 563 153
Total payments (in 109 stuyvers)
For half years 1- 99....28.15
100-119 .... 8.91; 8.91 .2=5.9
120-139 .... 9.30; 9.30 • ½=4.6
140-153 .... 6.67; 6.67 • ½=2.2
Sum total . . . . . . . 40.9
Sum of chances . . . . . 128
Mean payment . . . . . 0.320
Notes. Data in column 2 taken from pp. 12-15 of the memorandum. On pp. 15 16 of the memo-
randum the calculation is continued until half year 200. De boven staende Taeffel, as stated on p. 16,
was calculated by T. BEI~LECrIIEl~and JACOBLENSE,yder apart. On pp. 16-20 the same data, though
only until half year 153, is entered in DE WITT'S main table. The calculation of total payment by
DE Witt himself is not rounded off as above.

DE WITT'S m a i n s u p p o s i t i o n s r e g a r d i n g m o r t a l i t y are ([45], p. 234) :

the likelihood or chance o f dying in a given year or half year ... from 53 to 63 years
o f his (the a n n u i t a n t ' s ) age, taken inclusively, does not exceed more than in the
proportion o f 3 to 2 the likelihood or chance o f dying in a given year or half year
during the ... vigorous period of life ( f r o m t h r e e to 53) ... D u r i n g age 6 3 - 7 3 this
c h a n c e cannot be estimated at more than double .... and as the triple ... during the
7following years.
T h e s e c h a n c e s are e n t e r e d in the first c o l u m n o f DE WITT'S m a i n t a b l e ([100],
pp. 1 6 - 2 0 ; see m y T a b l e 2.3.3(1)) in w h i c h the s e c o n d c o l u m n shows h o w m u c h
the a n n u i t a n t will receive, a l l o w i n g for the d i s c o u n t , if o n l y he s u r v i v e s t h e cor-
r e s p o n d i n g n u m b e r o f h a l f years. T h e c o u n t o f h a l f y e a r s b e g i n s w i t h age t h r e e
a n d the d i s c o u n t m e a n s that, for e x a m p l e ,
9,805,807= 107:]/1.04
1 9 , 4 2 1 , 1 9 2 = 107: ~ + 107. (1]/-i~.04) z

4 7 9 , 8 2 0 , 5 6 3 = 10 v [1 : ] / 1 . 0 4 + 1 :( l]/-i~.04)z + ... + 1 :(1]/'i~.04)153] etc.

I n a n a p p e n d i x to the m a i n text o f the m e m o r a n d u m ([100], pp. 2 3 - 2 4 ; [45],

pp. 1 7 7 - 1 1 8 ) DE WITT a d d s :

... since the proof o f the foregoing demonstration I have had very carefully extracted
from the registers o f your Lordships some thousands o f eases o f persons upon whose
lives annuities have been purchased...
214 o . B . SHEYNIN

Examining considerably more than a hundred different classes, each class

consisting o f about one hundred persons, DE WITT found that
for young lives each o f these classes always produced to the annuitants ... a value o f
more than sixteen florins o f capital arising fi'om one florin o f annual rent . . . . Thus
one finds with wonderment, that in practice, when the purchaser o f several life
annuities comes to divide his capital.., upon several young lives--upon ten, twenty,
or more--this annuitant may be assured, without hazard or risk o f the enjoyment o f
an equivalent, in more than sixteen times the rent which he purchases.

Noticing that it remains impossible to check the correctness of DE WITT'S use

of his data, HENDRIKS ([45], p. 117) is satisfied that

experimental observations and collection o f the indications o f mortality at various

ages, and in different classes o f lives, were the principles in which the acuteness o f
De Witt and Hudde (who officially declared his agreement with the principles of
DE WITT'S calculation) recognized the only true foundations for their labours.

DE WITT'S statement about the effect of combining several annuities is related

to the general idea which underlies the law of large numbers 1. In the opinion of
J. BERNOULLI([3], chap. 4 of part 4) the qualitative form of this law was generally
known, though possibly in life rather than in science. See however §2.4.4. It is
indeed regrettable that DE WITT'S calculations were not found. In particular, it
would have been interesting to see just how he arranged annuitants into classes.
DE WITT'S exposition is ([26], p. 66)

trOs obscure, on l'a ordinairement mal comprise (an example of a misunderstanding

by M. CANTOR follows). De Witt parle du risque de mourir, sans fair ressortir
expressOment que ce risque se rapporte toujours dun enfant ayant 3 ans; au contraire
il s'exprime de maniOre d faire croire que le risque de mourir entre x et x + 1 ans
accomplis signifie ce risque pour une personne ayant actuellement x ans. De plus,
il fait deux diffOrentes suppositions sur les nombres des dOcOs semestriels ... dans
l'exposition ... il suppose ces nombres proportionnels d 1, 2, 2, 3, mais dans l'appli-
cation de la mOthode iI introduit, sans un mot d'explication, l'hypothOse in which
the corresponding numbers are 1, ~-,2½and ½.

To prove this fact ENESTR6M adduces tables of mortality compiled under each of
these hypotheses and concludes (p. 68) :

En examinant ces deux tables, on voit ... que la derniOre ... indique une mortaIitO
beaucoup plus grande que la premiOre. On pourrait donc conjecturer que De Witt,
apr~s s'~tre servi de la premiOre hypothOse pour calculer la valeur de la rente viagOre,,
eFtt trouvO cette valeur trop grande, et que, pour cette raison, il choisissait une autre
hypothOse (the second one) qui permettrait de vendre la rente gl un prix moins dlevO 2.

1 In this instance, it is the POISSON'Scase of the law of large numbers.

2 ENESTR6Mobviously did not know that HENDRIKS([45], p. 246) had pointed out the actual
assumption of DEWITT. However, HENDRIKSdid not say explicitly that DE WITTchanged his initial
hypothesis.
 Early History of Probability 215

Table 2.3.3 (2) which follows shows the difference between the two hypotheses
as noticed by ENESTR6M [26].
Table 2.3.3 (2)
Mortality According to DEWITT'S Declared (Hypothesis 1) and Actual (Hypothesis 2) Assumptions

No. of Half years Hypothesis 1 Hypothesis 2

interval
Chances of Annual Chances of Annual
death deaths death deaths

1 1-100 1 2 1 2
2 101-119 ~ 3 ~
3 120-139 2 4 ½ 1
4 140-153 3 6 ½

The simplest way to detect the discrepancy in DE WITT'S reasoning is to notice

that the chance to die during, say, interval No. 3 is simultaneously the chance to
survive intervals No. 1 and 2 (and not to survive interval No. 3). Thus the chances
of death should diminish as in hypothesis No. 2; as to the value of life annuities,
its calculation by DE WITT actually corresponds to this hypothesis because what
calculations demand is precisely these chances of death.
The mean duration of life and the value of life annuities according to hypo-
theses 1 and 2 are 46.4 and 33.2 years, and 18.8 and 16.0 years' purchase cor-
respondingly. Thus it was the mean duration of life rather than the value of
annuities which compelled DE WITT to change from hypothesis 1 to hypothesis
2. Alternatively, he (and HUDDE) could have overlooked the discrepancy al-
together.
Be that as it may, DE WITT'S hypothesis 2 led to an expectation of life which
coincided with the data given by STRUYCK([94], p. 212): in the average, purchasers
of annuities sold in Holland in 1672 lived to enjoy approximately 32 annual
payments. STRU¥CK'S data related to 1,698 annuitants, men and women, and he
gave figures for those under and above twenty for each sex separately. Thus, for
men (women) above twenty the mean expectation of life was 213 and 24½ years.
However, because DE WITa" assumed a single value of annuities for purchasers
of any age, this value could have been just only in the average; in particular, this
value was not just either for those above twenty (see Sa'Rt:YCK'S figures above),
or for purchasers of annuities on the lives of infants under three years of age, i.e.
on those not yet in the vigorous period of life.
In 1671, in a letter to HUDDE, DE Wnq" ([45], p. 109) explicitly formulated the
problem of calculating the value of annuities on several lives ~. Suppose, says
DE WITT,that eight persons have lived x~, x2, ... , x7, x 8 years, x 1 < x 2 < ... < x 7 < x 8 ,
correspondingly. (Actually his is a numerical example.) Then combinations of the
type (x i, x j) with i, j = 1, 2 . . . . . 7, 8 and i < j will constitute all possible cases of
duration of lives of two persons out of eight so that in one case out of Ca2 = 28 the
last survivor will live to age x2, in two cases, to age x 3 etc. Thus, in other words,
the law of longevity of the last survivor is obtained. Similar calculations with the

' The problem is possibly due to HUDDE; at least, DE WITT acknowledges a letter from HUDDE
on the same problem though offers no comment. See also below.
216 O.B. SHEYNIN

use of combinations of the type (x~, x j, xk) with i, j, k = 1, 2,..., 7, 8 and i < j < k
are used for the case of three lives etc.
DE WITT'S is only a methodological example; nevertheless, it should be noticed
that he determined the distribution of one of the order statistics (of the maximal
term of a sample). 1
The same correspondence explains why D~ WITT'S m e m o r a n d u m remained
for a long time a closely guarded secret.
I have properly understood the estimation o f the value o f life annuities upon one
life, computed (by w h o m ?)fi'om the life and death o f 96 persons, DE WITT begins
his letter of 2 Aug. 1671 ([45], p. 101).
Asking how to compute annuities on two lives, he continues (p. 102):
I leave you to consider whether ... you do not judge it to be useful for the public
good that this estimate should be absolutely hidden ... f o r the advantage o f the
State finances ...
Nevertheless, the fact of the existence of DB WITT'S m e m o r a n d u m had become
known to contemporary scholars while at least one of them (LEIBNIZ)actually
saw it. Thus in his letter to LEIBNIZ dated 20 April 1704 J. BERNOULLI ([33], p.
406) wrote :
Ich entnehme Deiner Beschreibung, daft die ... Abhandlung des Johann de Witt
Dinge enthiilt, die meinem Z w e c k besonders dienen. Ich bitte Dich deshalb, mir Dein
Exemplar des Buches sobald wie m6glich leihweise zu iiberlassen, da ich es eben im
Amsterdam vergeblich aufzutreiben versucht habe.
In another letter dated 28 Febr. 1705 J. BERNOULLI (ibidem) repeats his
request. Obviously, his Z w e c k was to finish his Ars conjectandi. However, he died
before doing so, and even without having seen DE WITT'S work, which opened
up new grounds for probability.

2.4. Political arithmetic and demographic statistics

2.4.1. Ancient and antique history. Statistical data including data on population
had been collected both in ancient times and later. I shall present the most im-
portant facts on this subject drawing on FEDOROVITCH([27], pp. 7--21). 2
In China, at approximately 2238 B.C., a geographical description of the terri-
tory and an estimation of the population had been attempted. It is possible that
censuses, including compilation of tables of population m o v e m e n t as well as
land registers were known in Egypt from the 35 th century B.C. The first census,
positively known to have been taken in Egypt, was carried out in the 16 th century
B.C. It covered only the warrior caste. Enumerations of the people of Israel are
described in the Old Testament. In Sparta LYCURGUS distributed land a m o n g his
subjects. This would have proved impossible without knowledge of the population.
1 Relevant achievements due to HUYGENS(§ 4.2.3) remained also unpublished. In another letter,
also in 1671 ([45], p. 102) DEWITTinforms HUDDEthat the number which you (HUDDE)have recently
proposed respecting the ratio between the chances of two players at quinque novem agrees literally with
the one which I had already calculated... A version of this game not devoid of interest is described
by MONTMORT([69], p. 173).
2 This is a well documented source. See also JOHN([53], pp. 17--34), MEITZEN([68], §§ 2 4) and
ELSNER[24].
 Early History of Probability 217

In Athens registers of births and deaths were kept; censuses were carried out
under PERICLES and SOLON. Freeborn citizens were entered upon special lists,
being registered at birth and also after reaching the ages of 18 and 20. In general,
registration of the population movement was facilitated by traditional offerings
to gods on the occasion of birth and death. In Rome lists of citizens able to bear
arms and Romans enjoying full rights, i.e. males aged 20 and 30, respectively,
were maintained. A table of life expectancies for various ages was compiled by
ULPIANUS (§2.3.2). A comprehensive description of Greek states and cities was
contained in ARISTOTLE'S Politica.
The general conclusion ([27], p. 15) is that
It could be hardly possible to agree with those who maintain that statistics
(I would say Staatswissenschaft) as a science was completely unknown to the
ancients and that it was founded only by Conring and Achenwall.
FEDOROVITCH (ibidem, p. /7) refers also to CONRING himself, who named
ARISTOTLE, STRABOand PTOLEMYas cofounders of the Staatswissenschaft.
This statement does not contradict the opinion of KENDALL([55], p. 45) that
statistics as we now understand the term did not commence until the seventeenth
century, and then not in the field of 'statistics'( ( = n o t in the Staatswissenschaft )
but in that of political arithmetic. The feudal state of the Middle Ages was just not
interested in statistics (in our sense),
KENDALLcontinues: even in 15th century Italy for all its achievements in account-
ancy and mathematics (p. 46)
counting was by complete enumeration and still tended to be a record of a situation
rather than a basis for estimation or prediction in an expanding economy.
I shall conclude by remarking that the statistical data collected for example
in Rome or Athens (see also §2.3.2) could have helped in the emergence of some
elements of political arithmetic. This possibility did not materialize owing to
general laws of development of ancient and antique civilizations which had no
need for demographic statistics as emerged during the 17th and 18t h centuries.
2.4.2. Petty. The term political arithmetic was introduced by W. PETTY, the
father of classic political economy. Describing states and separate cities from a
socioeconomical point of view in his Political arithmetic, PETTY ([79], p. 244)
rejected the use of comparative and superlative Words and took the course of
expressing himself in Terms of Number, Weight, or Measure; of using only Argu-
ments of Sense, and of considering only such Causes, as have visible Foundations
in Nature ...1
1 In one of his letters dated 1685 PETTY ([82], pp. 157-- 159) censures PASCALfor using many words,
phrases, and sentences.., which have no certain, sensible signification, and therefore cannot beget any
clear notion, sense, or science in the Reader.
The editor seems to prove that PETTY bears in mind PASCAL'SDiffOrence entre l'esprit de GOomOtrie
et l'esprit de finesse. In another letter dated 1667 PETTY ([81 ], vol. 2, p. 22) beg(s) leave of the worm
to decline the words Infinite, Eternall, Incomprehensible when (speaking) o f Almighty God. These
words, continues PETTY, are not soe fitt f o r Ratiocination, but rather for Adoration, they do not eleare
or brighten our understanding.
Still, people do use words and I have elsewhere noticed ([93], p. 109, ftn. 55) recent attempts
to quantify qualitative characteristics of m e n ' s social behaviour. A related topic is of course rank
correlation.
218 O.B. SHEYNIN

PETTY ([81 ], vol. 1, pp. 171-172) even proposed to establish a Register generall
of people, plantations & trade of England. In particular, he thought o f collecting
the accounts of all the
Births, Mariages, Burialls ... of the Herths, and Houses ... as also of the People,
by their Age, Sex, Trade, Titles, and Office.
As noticed by GREENWOOD ([39], p. 61) the scope of this Register was to be wider
than that of our existing General Register Office.
Strictly speaking, neither PETTY, nor, as it seems, his followers ever introduced
a definition of political arithmetic. However, without violating P~TTY'S thoughts
quoted above, it is possible to say that the aims of this new scientific discipline
were to study, from a socio-economical point of view, states and separate cities
(or regions) by means of (rather unreliable) statistical data concerning population,
industry, agriculture, commerce etc.
Thus PETTY ([80], p. 108) estimates the wealth of England determining the
value of
Housing, Shipping, Stock of Cartel, Coined Gold and Silver, Land as well as
the value of the people . . . . the Stock of the Kingdom yielding but 15 Millions of
proceed, is worth 250M then the People who yield25, are worth 416 2 M.
It is reasonable to object to the precondition of this calculation, i.e. to equating
population with wealth. More important, though, this calculation conformed to
the principles of statistics, the science which superseded political arithmetic, in
that the labour of each man and woman was estimated on the same footing.
At least until the middle of the 19th century the most interesting problems of
political arithmetic properly belonged to demographic statistics, so this branch
of political arithmetic came to the fore then. From my point of view, most import-
ant within demographic statistics were problems in mortality because of their
application to life insurance and, more specifically, to the institution of annuities
which emerged independently of political arithmetic (see §§2.3.2 and 2.3.3). The
founders of political arithmetic (see also §2.4.3) did not, of course, foresee the
importance of life insurance either for society or for mathematics (probability) 1.
Any detailed description of the work of PETTY in political arithmetic [39] is
out of the question and I shall offer but a few comments. Among the PETTY papers
cited [81] at least thirty pertain to political arithmetic. One manuscript ([81],
vol. 2, pp. 10-15, quotation below from p. 15; see also letter to E. SOUTHWELL
dated 1687, [82], pp. 318-322) is devoted to algebra which
came out of Arabia by the Moores into Spaine and from thence hither, and WP
hath applied it to other than purely mathematicall matters, viz : to policy by the
name of Politicall Arithmetick, by reducing many terms of matter to termes of
number, weight, and measure, in order to be handled Mathematically.
Algebra, maintains PETTY (p. 10) is a kind of Logick; in algebra (p. 14)
(1) The Algorithme is the Tooles. (2) The stock of axiomes is the Materialls
(3) The practice and a good head is the workmanship. (4) The finding out abstruse

1 However, LEIBNIZ,though not a cofounder of political arithmetic, regarded insurance of life

and property as a highlyimportant social institution [110], [116].
 Early History of Probability 219

truths is the work, and out of a few truths to draw out infinite true conc[lusions] and
to preserve the method of numbering unconfounded is the exc[ell]ency.
This subordination of algebra to logic as also PETTY'S meditations on Funda-
mentall questions show him as a philosopher of science possibly congenial in some
respects with LEIBNIZ, his junior contemporary. Indeed, among these Questions
([81], vol. 2, pp. 39-42) which PETTY asked himself are such as (pp. 39-40)
What is a common measure of Time, Space, Weight, & motion ? What number
of Elementall sounds or letters, will ... make a speech or language? How to give
names to names, and how to adde and substract sensata, & to ballance the weight
and power of words; which is Logick & reason. 1
The title of the second chapter of the Questions (p. 40) is What rules of Marriage
are best for procreation ? Here, as also in other manuscripts, PETTY advocates
improvements of biological conditions for the multiplication of mankind. The
same subject was treated in PETTY'S correspondence in 1685 [82]. Thus (pp. 148
and 154)
(1) It is for the honor of God and the advantage of mankind that the world
should be fully and speedily peopled, and that objections against the same may be
deferred till a thousand years hence (!)
That the more People there are in any Country, the greater is the value of each
of them ...
(2) ... till we see the Earth peopled (as perhaps 3 is not), wee may doubt (that
the whole Earth; and the fixed stars too, was made for the use of man); and not
knowing to what other use it was designed, may stumble into the Error of Its having
been made by Chance, and not by the designe of an Infinite wisdome ...2
So PETTY does use the word infinite after all! He does not forget to add (p. 155)
that the
King of England hath a greater share of the unpeopled Earth ... than most other
Princes (of prominent European nations?); wherefore when the whole shall bee
peopled, Hee will have a greater share than he hath now.
Multiplication of mankind continued to be the favourite subject of scholars
of the 18th century (e.g., of Siissmilch) .s
1 Cf. PETTY'S pronouncement ([78], p. 15): 1. Place is the Image or Fancy o f Matter or Matter
considered. 2. Quantity, the Fancy of Place .... 5. Situation, several Places considered together. 6. Figure
is Quantity and Situation considered together . . . . 9. Time, the Image of Motion. In this source (pp.
82-88) PETTY alleges that
(1) The likelihoods of reaching 70 years of age for those aged 16 and a (a < 16) are as ] f ~ - : ~ - .
(2) The likelihood of A aged a dying before B aged b (a, b > 16) is to the likelihood of the converse
event as Ira-: ]~-.
He does not refer to GRAUNT whose table of mortality (§ 2.4.3) contains nothing to corroborate
these conclusions. But then, PETTY illustrates his "laws" by examples mostly pertaining to men aged
16, 26 and 36 years, i.e. to ages which directly enter GRAUNT'S table.
2 For utterances on chance origin of the world see my earlier contribution ([93], pp. 134 and 140)
and also § 4.3.
a Is it possible that early eugenists, beginning with GALTON, saw any connection between them-
selves and S/3SSMILCH, if not PETTY? D. MAcKENzIE, in a private communication, informs me that
the answer to this question seems to be negative.
220 O . B . SHEYNIN

Two more sources from PETTY, though unrelated to political arithmetic,

deserve to be discussed. The first of these ([81], vol. 2, pp. 8-9) is obviously a
programme for educating and training young men. It mentions algebra (and, in
this connection, a number of scholars from VI~TE to WALLIS), geometry, chess,
games of chance and even hunting and fishing.
The second is a passage on lotteries ([77], p. 64) largely repeated by CONDORCET
and LAPLACE([91], p. 320):
A Lottery ... is properly a Tax upon unfortunate self-conceited fools ... Now
because the worm abounds with this kinde o f fools, it is not fit that every man that
will, may cheat every man that wouM be cheated; but it is rather ordained, that the
Sovereign should have the Guardianship o f these fools... (and monopolize lotteries ?).
Lotteries, continues PETTY, should be carried out but for small Leavies ...
2.4.3. Graunt. The cofounder of political arithmetic is J. GRAUNT. For a
long time his Observations [36] were attributed to PETTY. However, according to
HULL ([76], VO1. 1, p. lii) PETTY
perhaps suggested the subject o f the inquiry .... probably assisted with comments
upon medical and other questions here and there ... procured (some)figures ... and
may have revised, or even written the Conclusion ...1
In this book GRAUNT was able to use such fragmentary statistical data as
existed in his time, arriving at general quantitative estimates of the population
of London and England and of the influence of various diseases on mortality;
see also § 2.4.5. In particular, according to chap. 8, the difference between London's
male and female populations equalled about a thirteenth of the latter. As GRAUNT
remarks, because comparatively more men die violent deaths, travel, and remain
unmarried such as
fellows of colleges, and apprentices above eighteen etc . . . . the said ... difference
bringeth the business but to such a pass, that every woman may have a husband,
without the allowance o f polygamy.

In chapter 11 GRAUNT estimates the population of London. Here is one of

the methods he employs (pp. 68-69) :

I guessed that in 100 yards square there might be about 54 Families ... There are
220 such squares within the Walls (11,880 families). But forasmuch as there dy
within the Walls about 3200 per Annum, and in the whole about 13 000 ... the whole
population of London consists of 47,520 families.

GRAUNT'S celebrated table of mortality (see Table 2.4.3 (1)) compiled from
bills of mortality is in the same chapter (p. 69). Information about the age of deaths
was almost completely lacking in these bills, but G~UNT'S statistical insight and
ingenuity enabled him (or, rather, PETTY; see below) to compile this table draw-

1 If so, PETTY still seems to be the coauthor of the Observations. To add m y own bit to this problem
of authorship I quote from PETTY'S address to Lord BROUNKER [78] : I have also (like the Author of
those Observations [on the Bills of Mortality]) Dedicated this Discourse to... the Duke of Newcastle...
 Early History of Probability 221

Table 2.4.3 (1)

GRAUNT'S Mortality Table ([36], p. 69)

Of 100 there dies

within the first six years . . . . . . . . . 36 The fourth . . . . . . . . . . 6
The next ten years, or Decad . . . . 24 The next . . . . . . . . . . . 4
The second Decad . . . . . . . . . . . . . 15 The next . . . . . . . . . . . 3
The third Decad . . . . . . . . . . . . . . . 09 The next . . . . . . . . . . . 2
The next . . . . . . . . . . . 1

Note. On pp. 69-70 GRAUNT appends a table of survivors. For example, the entry which corres-
ponds to age sixteen in the new table is 4 0 [ = 1 0 0 - ( 3 6 + 2 4 ) ] etc.

ing on his own conclusions about mortality from various diseases, childrens'
diseases included. GRAUNT'S own explanation of the method he used (p. 69) is
generally known and to this method the opinion of WILLCOX ([36], p. x) part-
icularly applies:

To the trained reader Graunt writes statistical music; Petty is" like a chiM playing
with a new musical toy which occasionally yields a bit of harmony ...1
And, on p. xiii:
Graunt is memorable mainly because he discovered ... the uniformity and
predictability of many biological phenomena taken in the mass ... thus he, more than
any more man, was the founder o f statistics.
In this context, uniformity and predictability mean that final results of statistical
inquiries (such as the mortality table) could be used for a certain number of years
to come, a fact implicitly supposed by GRAUNT (see below).
WILLCOX([36], p. xi) opines that the table itself is due to PETTYwho incidentally
and characteristically ignored GRAUNT'S theory that seven percent survived
seventy 2 ... assuming instead, without reason, that one percent survived seventy-six
and not one percent eighty-six ...
As I understand him, WILLCOXalso supposes that PETTY calculated a constant
chance (p) of dying by decades from the equation

64 (1 _p)7 = 1 (2.4.3.1)
in which 64 and 1 are the numbers of people alive at ages 6 and 76, respectively.
The same opinion is held by HACKING ([43], p. 109) who continues: The solution
f o r p is very nearly 3 (so that 1 - p =-~). Now this value (5) explains the composition
of GRAUNT'S table of survivors but the solution of (2.4.2.1) really is p = 0.4479 4:3 !

1 As HULL noticed ([76], vol. 1, p. lii) PETTY sought to consider even the number of sea-fish
and wild-fowl at the end of every thousand years since the Flood. No wonder his statistical estimates
were often wrong; even so, it was he who first advocated the use of the new toy and (see above) sug-
gested the subject of GRAUNT'S inquiry. For a vivid characteristic of PETTY see also GREENWOOD
([371, p. 80; [391, p. 73).
2 On this occasion GRAUNT'S statistical music was written in a rather disorderly fashion; this
supposition is found elsewhere ([36], p. 32). As to PETTY'S authorship, see also § 2.4.2 for a description
of a related (and unfounded) study due to PETTY.
222 O.B. SHEYN1N

In 1937, PTOUKHA([83], p. 71) reconstructed GRAUNT'S table assuming

l-p= 6 4 - 1 =0.63 5.
100 8
However, this method of determining p is of course completely arbitrary.
GREENWOOD ([39], p. 79) noticed that, according to GRAUNT, later age
mortality is enormously higher than according to HALLEY.Granted. But why does
GREENWOOD infer that this shot did not find the bull's eye? ... GRAUNT must be
credited for inventing the table if not for its accuracy (unattainable in his time).
GgAUNT'S table, as he himself said ([36], p. 70) enables one to estimate the number
of fighting men. Other eventual uses of the inquiry as a whole are mentioned on
pp. 78-79, in the Conclusion: the
Art of Governing, and the true Politiques, is how to preserve the Subject in
Peace, and Plenty ... Now, the Foundation, or Elements of this honest harmless
Policy is to understand the Land, and the hands of the territory to be governed ...
It is no less necessary to know how many People there be o f each Sex, State, Age,
Religion, Trade, Rank, or Degree, etc. by the knowledge whereof Trade, and Governe-
ment may be made more certain, and Regular ... whether the knowledge thereof
be necessary to many, or fit for others, than the Sovereign, and his chief Ministers,
I leave to consideration.
The last remark shows that even GgAUNT (or PETTY?) did not yet recognize
the importance of statistics for a broad circle of educated citizens.
I shall conclude with three estimates of G~UNT'S work. The first is found
in HUYGENS' letter dated 1662 ([47], t. 4, p. 149); the others ([83], p. 67; [34],
pp. 13 and 14) belong almost to our time.
(1) Le discours de Grant (!) est tres digne de consideration et me plait fort,
il raisonne bien et nettement et j'admire comment il s'est avis~ de tirer toutes ces
consequences hors de ces simples observations, qui jusqu'a luy ne semblent avoir
servi de rien. dans ce pais icy l'on n 'en fait point, quoy qu'il seroit a souhaiter qu'on
eust cette curiositO et que la chose soit assez aisle, principalement dans la ville
d'Amsterdam, qui est tout divisOe en quartiers, et dans chascun il y a des prefects
qui scavent le nombre des personnes et tout ce qui s'y passe.
(2) The trois plus grands m&ites de Graunt devant la science statistique are:
1° ilfut le premier qui ~tablit d'apr~s des mat~riaux statistiques les lois empiriques
qui sont propres aux phgnomOnes collectifs atypiques; 2 ° il montra la maniOre
pratique avee laquelle on peut et on doit utiliser les donnOes statistiques aprOs les
avoir soumises ~l une analyse critique; 3 ° il Otablit la premiOre table de mortalitO.
(3) ...the concept of a life table was an outstanding innovation and it lay ready
for Halley's use...Graunt's work created the subject of demography. (It) contributed
to statistics in general.
Even if GRAUNTdid not actually create the subject of demography completely
alone, he at least published the first quantitative demographic research.

2.4.4. Leibniz. As to probability proper, LEIBNIZ should be credited with

findings mainly in the field of games of chance; they have been described in a
 Early History of Probability 223

series of articles by K.-R. BIERMANN and K.-R. BIERMANN ~,~ MARGOT FAAK.
Besides this, LEmN~Z claimed there was need for a probability logic ([93]), p. 115) ~.
Lastly, the correspondence between LEIBNIZ and J. BERNOULLIwas also extremely
valuable.
Part 4 of the Ars conjectandi was written with a mind to LEmNIZ' opinion.
BERNOULLI had confided to LEIBNIZ his ideas about using statistical probability
on a par with theoretical probability. LEIBNIZ, at least initially, disagreed. As
I said before ([93], p. 138), he may have been prepared to weigh delicate subjective
opinions and probabilities rather than enumerate successful and unsuccessful
trials.
I would now state more definitely that, according to LEIBNIZ ([109], p. 288
and letter to J. BERNOULLI 3 Dec. 1703, [33], p. 405),
(1) ... deren [zufiillige Dingen are meant] vollkommener Beweis jeden endlichen
Verstand iiberschreitet.
(2) Was yon unendlich vielen Umstdnden abhdngt, nicht durch endlich viele
Versuche bestimrnt werden kann...
This reservation is possibly borrowed from the Logique de Port-Royal ([2],
p. 372) :
(1) Il est de la nature d'un esprit fini de ne pouvoir comprendre l'infini.
(2)...ce serait un ddfaut de raison de s'imaginer que notre esprit Otant fini, il pftt
comprendre jusqu'ofi peut aller puissance de Dieu, qui est infinie...
In any case, it proved pessimistic: science generally, and mathematical
statistics in particular, has merely to do with transitions from finite to infinite.
But of course a final confirmation of facts or hypotheses by statistical data is
theoretically impossible: in this sense LE~BNIZ is absolutely right.
My last comment is that, in accord with his general philosophical point of
view, LEIBNIZ likely believed in deduction rather than in induction; see however
the discussion of his contribution [62] below. Only in 1714, in a letter to one of
his correspondents (I-112], p. 570) he seemed to recognize the principle of post-
erior estimation of probability:
On estime encore les vraisemblances a posteriori, par l'experience, et on y doit
avoir recours au ddfaut des raisons a priori: par exemple, il est egaIement vrai-
semblable que l'enfant qui doit naistre soit gaff on ou fille, parce que le nombre des
garfons et des filles se trouve d peu pros egal dans ce Monde. L'on peut dire que ce
qui se fait le plus ou le moins est aussi le plus ou le moins faisable dans l'etat present
des choses, rnettant toutes les considerations ensemble qui doivent concourir d la
production d'un fait.
Considering the development of the theory of probability LEmNIZ also says,
just before the passage quoted:
Feu M. Bernoulli a cultivO cette matiere sur mes exhortations.
1 A discussion of LEIBNIZ'work on the subject is beyond my purpose. Readers may look up the
book by HACKING[43] with referencesto his own previous articles as well as other commentators,
BIERMAN~and BIERMAtCN& FAAKincluded.
224 O.B. SHEYNIN

This I fail to understand: in a letter dated 3 Oct. 1703 BERNOULLIwrote to LEIBNIZ

([33], p. 404):
Ick mdckte gerne wissen, grosser Meister, yon wem Du erfahren hast, dass ick
mich mit der Lehre yon der Abschiitzung der Wahrscheinlichkeiten beschdftigt
kabe.
I shall now turn to political arithmetic. This new discipline caught LEIBNIZ'
imagination, the more so because his and PETTY'S attitudes towards philosophy
of science were similar (§ 2.4.2). LEIBNIZ' writings on political arithmetic [60 ]-[64],
all unpublished during his lifetime, had been collected by KLOt'I' [59] who main-
tains (p. xxxviii) that the first three of them
in die achtziger Jahrefallen and that LEIBNIZ' Essay [63] was called forth by PETTY'S
Political Arithmetick about which LEIBNIZ also kat...eine Menge einzelner Be-
merkungen und kleiner Aufsdtze niedergesckrieben.
One such remark found by KLOPP auf einem losen kleinen Blatte is quoted by
him (p. xxxvii) :
Il faut reduire toutes les sciences en Figures et en Formules." ear, plusieurs
ckoses ne pouvant estre exprimOes par figures (si non analogiquement ce qui n'est
pas scientifique), pourront estre au moins assujetties aux formules qui tiennent
lieu de figures, et servent ?l arrester l'imagination...
LEIBNIZ advocated the compilation of Staatstafeln ([60], p. 303):
Ich kenne Staats-Tafeln eine schriftliche kurze Velfafiung des Kerns aIler zu
der Landes-Regierung gehdrigen Nachrichtungen, so ein gewifies Land insonderkeit
betreffen, mit solcken vortheil eingerichtet, daft der Hohe Landes-Herr alles darin
leicht finden, was er be), jeder begebenheit zu betrachten, und sick defien als eines
der bequdmsten instrumente zu einer 16blichen selbst-Regierung bedienen k6nne.
He noted the benefit of comparing such tables with each other ([61], p. 316):
Sonderlick aber wiirde grofien Nuzen kaben die comparation untersckiedener
Lande, orthe unter einer herrschafft, und unterschiedener Zeiten eines landes gegen
einander .... Daraus dann allerkand reale verbefierungen erfolgen wiirden...1
LEtBNIZ (ibidem, p. 317) also mentions special- und general-Registraturen the
latter possibly being the same, or of the same nature, as the Staatstafeln. He
urges (ibidem, p. 319) the need to follow the example of theologians and lawyers:
Die Tkeologi haben Harmonias confessionum; die juristen haben Differentias
variorum jurium.., weit niizlicher wi~rde sein eine Harmoni und collation in Regie-
rungssacken, dadurch der Herrschafft und gemeinen Wesen viel nuz zu schaffen,
welches alles zu diesem Registraturen Amt eigentlick gek6hret.
Thus LEIBNIZ actually recommends establishment of a special Amt 2 and a list

1 Unification of national statistical data, a problem tackled by statisticians in the second half
of the 19t~ century, proved extremely difficult.
2 A pertinent remark is due to COUTURAT([18], p. 522): In 1704 LEmNIZpensa...~fonder une
SociOtd des Sciences...?tDresde. One of the aims of this society in his opinion would have been to
dresser des statistiques dOmograpkiques. See also BIEDERMANN([5], p. 457).
 Early History of Probability 225

of fifty-six questions [64] seems to correspond to a programme, or a part of a

programme, for its statistical inquiries. Among these questions are such as
1. Numerus hominum.
4. Quot foeminae aptae ad generandum.
21. Comparatio mortium et nativitatum.
26. Quanta sit agri frugiferi media foecunditas intra 7 circiter annos.
47. Descriptio exacta omnium artium et vitae professionum. 1
L~I~N~z [62] gives special attention to medicine. In his opinion (p. 321)
...der juristen insgemein zu vieI, der Medicorum aber zu wenig seyen...
Considering also peoples' Blindheit, when (p.322)
es geht den meisten mit der gesundheit wie mit der seeligkeit, deren keines sie achten,
bis sie yon der spdthen reue iibereilet werden,
and, following his general line of thought, LEIBNIZ (p. 321) advises a

...zusammensetzung der bereits vorhandenen wiJ3ensehafften, Erfindungen, Experi-

menten und guther gedancken, which would bring under control vielen Kranck-
heiten.
He also urges practitioners to record their observations (p. 325):
...wenn jeder practicus nur einen einigen richtigen Aphorismum zu den Hippo-
craticis oder andern bereits bekandten gefftget hdtte, man jezo weit kommen sein
wiirde. Ieh nenne aber aphorismum nicht eine jede thesin, sondern diejenigen sdze
so nieht durch die vernunfft erhellen, noch yon selbsten sich verstehen, sondern aus
der etfahrung vermittelst fleifliger beobachtung entdecket werden. Wiewol derjenige
so ein seh6hnes theorema oder vernunfftschlufl, deJ3en man sich nicht leicht versehen
solte, durch scharfes nachsinnen a priori oder aus betrachtung der Ursachen auflge-
funden hdtte, so durch die erfahrung hernach richtig befunden wiirde, wegen solcher
seiner scharfsinnigkeit nicht weniger als jener wegen seiner fleiJ3igen aufmereksam-
keit zu loben und zu belohnen.

The last lines are related to the general problem of comparing deductive and
inductive methods.
On the whole, LEmMz' contribution [62] is in line with the ideas of political
arithmeticians concerning the betterment of human life and multiplication of
mankind (see § 2.4.2).
A special point made in LEIBNIZ' work [62] is a concrete proposal (p. 322)
to establish a Collegium Sanitatis so as to supervise shops, bakeries etc. More-
over (p. 323),

Die Acta und Archiva des Collegii Sanitatis k6ndten und miisten unter andern
in sich halten, was in gesundheitssachen, und damit verwandten Dingen yon Zeiten
zu Zeiten passiret, und sonderlich wie in diesen und benachbarten orthen das wetter
sich gewechselt ..... wie sich das gewicht der lufft auch des magnets declinationen

1 The questionsjust quoted (and some others) seem to be intelligibleeven for those who do not
read Latin, myselfincluded. But a thorough study of LEIBNIZ'questionnaire is still warranted. Feei
quod potui, faciant meliora potentes!
226 O.B. SHEYNIN

und inclinationen gegindert, und was dergleichen durch die neuen instrumenta, nehm-
lich Thermometra, Hygroscopia, Anemia, Barometra und gewifle Compasse zu
entdecken. Ferner wie diese oder jene orth yon friichten und obst gerathen, was die
victualien fiir einen preifl gehabt, fi~r allen Dingen aber was fiir Krankheiten und
Zuf~ille unter Mensch und Vieh regiret, da dann die Symptomata, auch juvantia
und nocentia (wholesome and harmful) sammt allen umbstdnden aufs genaueste
zu besehreiben.
I do not think any such Collegium ever came into existence! In his Essay LEIBNIZ
([63], p. 328) supposes that the
Bornes ordinaires de la vie humaine sfavoir 80 ans, comptant pour rien le petit
nombre de ceux qui les passent
and that (p. 329)
...81 enfans nouvellement n& mourront uniformement, c'est g~ dire g~ un par annOe
dans les 81 ans suivans.
He then (pp. 330-332) calculates the moyenne longueur de la vie humaine both for
newly born infants and people of any age, necessary, as he notes, for estimating
the value of life annuities. Assuming one more supposition (p. 334), viz,

que la fecunditO des hommes est aussi tousjours la m~me et teIlement egale ~ leur
mortalitO, he notes (ibidem) that the multitude des hommes ne change pas notable-
ment, si non par quelques accidens particuliers et extraordinaires.
Il s'en suit par l~, LEmNIZadds on p. 335, que si 100 enfans de dix ans meurent, il
mourront aussi lOO personnes de 20 ans, et lOO personnes de 30 ans, et generalement
autant d'un aage que d'un autre...car si les vieillards sont plus sujets naturellement
mourir, leur nombre aussi est plus petit ~ proportion...
One more conclusion (p. 336) is that II meurt ~ peu pros la quarantieme partie
des hommes par an. This conforms to experience, quoyque on l'ait trouvd a priori
et par le seul raisonnement.
I must also remark that L~tBNIZ (p. 327) introduces apparence which n'est
autre chose que le degrO de la probabilitO ; thus, for an ordinary die, l'apparence est
egale pour chacune de ces faces...
There also exists une Apparence Moyenne (expectation) an example of cal-
culating which follows (ibidem):
Supposons...qu'il s'agisse de sfavoir la valeur de quelque heritage, maison ou
autre bien, qu'on doit estimer...
The value sought, explains LEIBNIZ, is arrived at by trois bandes d'estimateurs;
see his Neue Abhandlungen fiber den menschlichen Verstand ([93], p. 110) where
the whole passage is repeated, the only essential difference being the change
from Apparence Moyenne to Prostapherese. Thus LEIBNIZseems to leave the term
probability beyond the realm of mathematics; however, he does not say so.
I shall now discuss the opening lines of the Essay (p. 326):
Cette recherche peut avoir un usage considerable dans la politique: l'un pour
juger de la force d'un estat, et du nombre des personnes vivantes par le nombre
 Early History of Probability 227

Table 2.4.4 (1)

Extract from HALLEY'Stable ([44], p. 6)
Age current Persons
1 1000
18 610
26 560
35 490
43 417

des morts qui se voit dans les listes des mortuaires, 1 qu'on a coustume de dresser
sur latin de chaque annde ; l'autre pour estimer la longueur moyenne de la vie d'une
personne, d tin de donner une juste valeur aux rentes d vie, qui sont d'une grande
utilitd dans l'estat, comme f e u Monsieur le pensionnaire de Wit a fair voir...
See also § 2.4.2 where I emphasize the importance o f the institution o f annuities.
As it remained unpublished, the Essay with its i m p o r t a n t if simple conclusions
did not influence such scholars as HALLEY (or, maybe, HUYGENS). In particular,
the possibility o f estimating populations overlooked by GRAUNT and u n d e r s t o o d
by LEmNIZ was (independently) pointed out by HALLEY (§ 2.4.5).
2.4.5. Halley. In 1694 HALLEY published a m e m o i r [44] which played an out-
standing role in the f o u n d a t i o n o f d e m o g r a p h i c statistics. 2
D r a w i n g on incomplete and inaccurate data on mortality in various age
groups he arrived at his main result, a mortality table, or, m o r e correctly, at a
table o f survivors, for a stationary population. A n extract f r o m his table is given
above (see Table 2.4.4(1)). Being unsatisfied with his initial data because of its
irregularity HALLEY attributed it to chance (p. 5):
...yet that (irregularity) seems rather to be owing to chance, as are also the
other irregularities in the series o f age, which would rectify themselves, were the
number o f years (of observation) much more considerable...
Moreover, using additional data on births, HALLEY actually rectified these
irregularities.
This line of action is not really convincing: adjustment o f data in d e m o g r a p h y
is mainly aimed at detecting (and excluding) systematic influences rather than
chance effects? However, HALLEY seems to have been the first to correct the
1 Actually LEIBNIZdid not say anything more about calculating populations.
2 Der Tag, an welchem E. Halley seine Abhandlung...vortrug, darfals der Geburtstag der statisti-
schen Wissenschaft bezeichnet werden.
This opinion of BOCKH([8], p. 1) is an overestimation: it seems impossible to speak about the
Geburtstag of statistical science without mentioning PETTYand GRAUNT.
3 Thus, discussing mortality from various diseases, GRAUNTfirst and foremost strove to correct
systematic influences; he reasonably supposed that the death-rate from syphilis was grossly under-
estimated because those who died of it were usually returned of ulcers etc. GgAUNT'Sbook ([36],
p. 39) also contains a curious passage on freaks of chance:
...the rickets were never more numerous than now, and...they are still increasing; for anno 1649,
there were but 190 (cases of death of rickets), next year 260, next after that 329 and so forwards, with
some little starting backwards in some years...
228 O.B. SHEYNIN

observed frequencies of events. As in the case of DE WITT (§ 2.3.3) HALLEY'S

statement is related to the general idea which underlies the law of large numbers.
HALLEYused his table for a number of stochastic considerations, in particular
for calculating probabilities concerning the lives of two men. One of his problems
is this: calculate the probability that two men, aged 18 and 35 correspondingly,
will both remain alive after eight years; that they will both die; that only one
of them will remain alive (two cases). Calculating these probabilities, HALLEY
actually used the multiplication theorem for independent events. For example,
the answer to the first question is (see Table 2.4.4 (1))

50.73
P- (2.4.4.1)
610 • 490

(50=610-560, 73=490-417).
It seems that HALLEY was not satisfied with such analytical procedures.
Possibly imitating ancient mathematicians, he repeated his considerations using
the geometrical method, even in the three-dimensional case with each number
from (2.4.4.1) and from similar expressions for other probabilities corresponding
to a length of a side of a certain rectangle.
Discussing "chances" 50 • 73 and 610 • 490 HALLEY did not introduce prob-
abilities 6@0or 4@0.Neither did he introduce geometrical probabilities" the lengths
of sides of the rectangles were represented by integers. However, HALLEY did
offer a geometrical illustration of classical chances.
He explicitly pointed out (pp. 6-7) that solution of such problems as described
above makes it possible to calculate the value of annuities on two or three lives, 1
the
proportion of men able to bear arms, the different degrees of mortality, or rather
vitality, in all ages,

and the probable duration of life (the term itself is not introduced).

By what has been said, is one of HALLEY'Sconclusions (p. 8), the price of insurance
of lives ought to be regulated.
What he actually means is the comparison and possible adjustment of the price
of insurance (annuity ?) for men of various ages.

1 A small appendage to the main memoir begins thus (ibidem, p. 19): What I gave you in my
former Discourse on these bills, was chiefly designed for the computation o f the Values of Annuities on
Lives .... This addition mainly devoted to political arithmetic contains pronouncements such as (p. 21)
...the Strength and Glory o f a King being in the multitude o f his Subjects ete ..... Celibacy ought to be
legally discouraged . . . . And those who have numerous Families o f Children to be countenanced and
encouraged by such Laws as the Jus trium Liberorum among the Romans. But especially, by an effectual
Care to provide for the Subsistence o f the Poor, by finding them Employments...
In 1693 HALLEY([44a], p. 232 with a reference to Biog. Brit. 1757, a source I did not see)produced
a paper wherein he shewed a Method o f computing the Value of Annuitys for one two or three lives...
which was ordered to be printed in the Transactions. However, no such paper is mentioned in the List
of Halley's published writings (ibidem).
 Early History of Probability 229

A d d i n g up the n u m b e r s o f survivors at each age s h o w n in his table, HALLEY

achieved his m o s t i m p o r t a n t success: he thus c a l c u l a t e d the w h o l e p o p u l a t i o n
studied. E x a c t l y this c a l c u l a t i o n p r o d u c e d the m a i n i m p r e s s i o n on HALLEY'S
c o n t e m p o r a r i e s ([8], pp. 1--2): at a t i m e when censuses were c o m p l e t e l y u n k n o w n ,
HALLEY h a d s h o w n the w a y to e s t i m a t e p o p u l a t i o n s f r o m d a t a on births a n d
d e a t h s ) See also § 2.4.4.
HALLEY'S table greatly influenced DE MOIVRE. H a v i n g t u r n e d his a t t e n t i o n
to annuities on lives DE MOIVRE ([22], p. 262) c o n s u l t e d HALLEY'S m e m o i r a n d

f o u n d that the Decrements o f Life, f o r considerable Intervals o f Time, were in

Arithmetic Progression ... 2
T h u s the u n i f o r m d i s t r i b u t i o n was p r o p e r l y i n t r o d u c e d into p r o b a b i l i t y ? In
the course o f time, however, it p r o v e d t o o p r i m i t i v e a n d was s u p e r s e d e d by o t h e r
d i s t r i b u t i o n s . H o w e v e r , at the h a n d s o f such m a s t e r s as DE MOIVRE the use o f
even this p r i m i t i v e d i s t r i b u t i o n lead to the f o r m u l a t i o n a n d s o l u t i o n o f i m p o r t a n t
stochastic p r o b l e m s .

2.4.6. Neumann. The d a t a which HALLEY used were collected by CASPAR

NEUMANN (1648--1715) in Breslau where ([40], p. 206, f o o t n o t e 1), b e g i n n i n g f r o m
1542, die N a m e n der Getauften, Getrauten und Gestorbenen in dem Kirchenbuche
zu Maria Magdalena, und 1569 zu Elisabeth were e n t e r e d in ein Buch. 4
A c c o r d i n g to this source (p. 10), after s p e n d i n g three years at Jena U n i v e r s i t y
N e u m a n n wurde 1670 ... z u m Magister der Philosophic promovirt ... A Prediger
since 1673 (p. 12), NEUMANN c o r r e s p o n d e d with LEIBNIZ f r o m 1689 (p. 13). In
1706, on LEIBNIZ' r e c o m m e n d a t i o n , NEUMANN was elected a m e m b e r o f the Kgl.
Societdt der Wissenschaften in Berlin.
NEUMANN (p. 204)

1 Almost a century after the publication of HALLEY'S memoir T. PAINE, in his Rights of men,
presented arguments for national welfare activities. Requiring an estimate of those above fifty, PAINE
([56], p. 106)
several times counted the persons (he) met in the streets of London...and (had) generally found
that the average is about one in sixteen or seventeen (who are older than fifty).
Commenting on this estimate and referring to a number of sources KRUSKAL& PIETERS[56] sug-
gest that the proportion of those above fifty should have been 17 % or, possibly, any percent between
13 and 20. It is extremely interesting that HALLEY'Stable although compiled for a different population
(and time) would have furnished a figure of 18%! Thus PAINEcould have arrived at a rather trust-
worthy result just by using an old table--a classical table, I would add.
2 In another place ([21], p. x) DE Molvm~returns to HALLEYhis very hearty Thanks for Instructive
Notions readily imparted...during an uninterrupted Friendship of five and Twenty years...
A few lines describing the friendship between the two scholars are due to HELEN M. WALKER
([98], p. 356).
3 The first introduction of the same distribution by N. BERNOULLI(§ 2.2) remained unnoticed.
At least, nobody referred to N. BERNOULHin this connection.
A related example is provided without substantiation by ELSNER([24], p. 136): Thomas Crom-
well, entfernt verwandt mit...Oliver Cromwell, der Lordkanzler yon Heinrich VIII (1509-1547), be-
fiehlt in England die systematisehe Aufzeichnung yon Geburten und Todesfdllen in Kirchenbiichern;
in der Mark Brandenburg schreibt dies kurz danach die Visitations- und Consistorialordnung yon 1573
bindend vor.
230 O.B. SHEYNIN

... war, wie es scheint, in Deutschland der Erste, welcher iiber die Zahlenverhiiltnisse
der jiihrlichen Geburten und Todesfdlle, zuniichst innerhalb der Grenzen seines
Wohnorts, zusammenhiingende Beobachtungen anstellte und zu allgemeinen Schliis-
sen beniitzte.
NEUMANN'S main work seems to be lost (p. 207; see also below):
Als die Herausgabe des ersten Bandes des der Miscellanea Berolinensia ... vor-
bereitet wurde, ward zwar Neumann yon Leibniz zu einem Beitrage aufgefordert,
doch schickte er damals nichts ein. Erst in J. 1713 iibersandte er ... eine Abhandlung
... De methodo periodica in Obss, meteorologicis adhibenda . . . . (Der zweite Band
erschien lange nach seinern Tode 1723). Jedenfalls war Schlesien lange vor seiner
Einverbindung in den preuflischen Staat durch ... Neumann in der Berliner K6nig-
lichen Societiit ... wiirdig vertreten.
C. WOLFF, NEUMANN'S pupil and friend (p. 208), selbst hat beijeder Gelegenheit
dankbar bekannt, wie viel er ... Neumann verdankte, doch ohne gerade seine Me-
rhode zu riihmen.
GUHRAUER appends eleven letters exchanged between LEIBNIZ and NEUMANN.
In one undated letter (p. 265) NEUMANN remarks that he vil Jahre lang meteorolo-
gical observations begriffen gewesen.
In another letter dated 1707 (p. 267) he discusses the influence of the m o o n
on the weather:
Das aber der Monde mit der Luft ihren Veriinderungen einige Verwandniifl
habe, muthmaJ3e ich, sei auch schon bei den Hebriiern geglaubt worden.
Die Observationes meteorologicae erfordern eine gewifle Theorie, says NEU-
MANN in yet another letter of the same year (p. 269), ohne welche imand anders sich
schwer zum observiren schicken wird. Ieh k6nte aber schon damit dinen. Was biJ3her
yon solchen Dingen geschriben ist; oder auch die Parisische Societiit in ihren
letzteren Actis hat ... das ist alles vil zu wenig.
In the last letter dated 1713 (pp. 272-273) NEUMANN mentions his writing on
meteorology:
Habe mir aber gegen das Ende des abgewichenen Jahres die Ehre genommen ...
einen kleinen Discours de methodo periodiea ... beigeleget. Weil nun dises leztere in
ein besonderes an Ew. Excellence haltendes paquet eingeschloflen gewesen, und
vielleicht in Berlin rn6chte sein ligen gebliben; ich aber doch nicht g e m wolte, daft
dise wenige Arbeit verlohren gehen, oder in fi'emde Hiinde gerathen solte ...
A small essay on NEUMANN written by F, COHN is in the b o o k by GRAETZER
[35]. 1 COHN does not substantiate his attribution to NEUMANN of an attempt
(p. 27)
durch statistische Ermittelungen zu erproben, ob den wirklich ein Zusammenhang
zwischen Geburt und Tod der Menschen und gewissen kabbalistischen Zahlen oder
dem Stande der Planeten naehweisbar sei.
1 GRAETZER(pp. 33--37) also appends a letter from NEUMANNto JUSTELL,Regis Magnae Brit-
tanicae Bibliothecario, written in 1692. Here NEUMANNinforms his correspondent about plans to
conduct magnetic observations.
 Early History of Probability 231

Possibly COHN had in mind a letter of NEUMANN to LEIBNIZ dated 1689 ([40],
pp. 263-264) where NEUMANN says:

Als nehme ich mir endlich die Freyheit, einige Abschrift yon den bisher gemachten
Reflexionibus iiber Leben und Tod bei denen in Breslau geborenen und gestorbenen
zu iiberreichen, wiewohl das, das gegenwdrtige zu Ende laufende 89ste Jahr noch
nicht hat k6nnen beigefiiget werden. Noch zur Zeit kann man freilich nicht sehen,
was eigentlich der Nutzen davon sein werde. 1 SolIte aber Gott das Leben so lange
fristen, daft man die Rechnungen etzlicher Jahre zusammen bringen k6nnte, oder
auch jemand in einer andern Stadt dergleichen Observationes machen ... so wiirden
als denn sch6ne Anmerkungen g6ttlicher Providenz iiber unser Leben und Tod,
Erhaltung und Vermehrung der Welt, und dergleichen mehr k6nnen gemacht, auch
vielerlei Aberglaube desto besser aus der Erfahrung widergelegt werden. Ich beklage
sehr oft, daft itzund fast die ganze gelehrte Welt in regno Naturae sich auf Experi-
menta leget und Observationes schreibt, aber kein Mensch dergleichen in regno
gratiae oder in der Theologia zu thun gedenket ...

For his part, LEIBNIZ, in a letter to JUSTELLdated 1692 ([65], p. 279) testified:

Mons. Neuman (!) ... a fait des bonnes remarques sur les mortuaires et bap-
t6mes de la ville (Breslau), qu'on m'a communiqudes. Entre autres il observe que
les contes des ann&s climacteriques ne se verifient point.

Most likely NEUMANN rejected the contes as a result of a c o m m o n sense analysis

of data. Still, for his time, this is something well worth remembering.
A letter from HALLEY to NEUMANN on mortality dated 1692 is published in
the original Latin in HALLEY'S Correspondence ([44a], pp. 88-89). A letter from
NEUMANN to HALLEY dated 1694 kept at the Royal Society (ibidem. p. 35) was
published by GRAETZER ([35], p. 42).

3. Pascal and Fermat

This section is mostly devoted to the correspondence of PASCAL and FERMAT,
initiated, as it seems, by the former. PASCAL'S attention had been attracted to
games of chance by DE MkR~, a man of the world well versed in them although
not a gambler by vocation ([73], p. 409).
The extant part of the correspondence between PASCAL and FERMAT ([29],
pp. 288--314 and 450-452) includes eight letters; two of them, written in 1660,
do not directly bear on my subject. The other six letters were written in 1654 and
games of chance are discussed in four of them. It is these letters that I shall de-
scribe below. I shall also mention in passing the fifth letter of the year (FERMAT-
PASCAL, 29 Aug. 1654).

Did NEUMANNknow anything about GRAUNT or PETTY.9 It is not clear whether the Reflexionibus
were ever published. ELSNER ([24], p. 138) names the titles of two of NEUMANN'S works sent by the
latter to HALLEY via LEIBNIZ: Schdne Anmerkungen g6ttlicher Providenz iiber unser Leben und Tod
and Reflexionen iiber Leben und Tod bei denen in Breslau Geborenen und Gestorbenen.
232 O.B. SHEYNIN

3.1. Fermat-Pascal, letter without date ([29], pp. 288-289)

Gambler A tries to accomplish a certain score with a single die throwing it
eight times in succession. Suppose now that A is to throw the die only seven times.
What part of the stake is due him as compensation?
Par mon principe, says FERMAT, obviously answering a letter from PASCAL,
A is to receive ~ of the stake. If A will throw the die only six times he is to receive
additionally ~ of the remainder, i.e. I ' ~ = ~ of the stake etc. Finally, if, for
example, the first three throws prove unsuccessful, A's compensation for one of
the remaining throws is still 61-of the stake.
The probability of scoring k points ( k = 1, 2,..., 5, 6) in eight throws exactly
once is less than the probability of the same event happening in seven throws;
generally, in n throws

Pn =-~- , Ps <P7.

For this reason and, of course, bearing in mind the general context of FERMAa"S
letter, I shall interpret the condition of the problem as scoring k points at least
once.
Denote the probability of success in each throw by p (p = ~). Then the prob-
ability of success in two throws is

p + p (1 - p ) = 2 p - p 2 (3.1.1)
and in three throws

p +p(1 - p ) + p {1 - [p +p(1 -P)I} = 3 p - 3p z +p3.

These formulae are rather simple. Much more interesting is the general formula
tl

P{2Ai}=2P{Ai}-2P{AiAj} + 2 {AiAjAk}-""
i=l i i<j i<j<k

known to DE MOIVRE and SIMPSON ( [ 9 0 ] , pp. 278-279).

Thus FERMAT used the "classical" definition of probability and, also, the
concept of mathematical expectation of a random event. The essence of FERMAT'S
principe, as I see it, was to calculate expected gains (or losses).
Neither FERMAT nor PASCAL formally introduced probability or expectation
nor did they introduce any formal term for expectation. Lastly, expressions like
(3.1.1) involving "pure" probabilities possibly remained foreign to both of them.

3.2. Pascal-Fermat, letter dated 29 July, 1654 ([29], pp. 289-298)

The letter begins by discussing the problem of points (3; 2:1), with notation
as in §2.1 above. A wins the fourth set (and the whole game as well) with prob-
ability ½. Therefore, argues PASCAL, a half of the stake is already due him. If,
however, A loses the fourth set, the score equalizes so that A is to receive ad-
ditionally ½ • 1 = ~1 of the stake. The total share of the stake due A is thus 3.
 Early History of Probability 233

3.2.1. Random walks. FREUDENTHAL ([31 ], chap. 7) interpreted this problem,

or rather the problem (5; 4:3) in terms of random processes. Point M1 (4; 3)
(see Fig. 3.2.1 (1)) corresponds to the score. As the game proceeds M 1 changes

M 1 (4; 3)
M2 (4;4)
M2 M3 M 3 (5; 3)
3~t3 (5; 4)
M 4 (4; 5)

M1 M3
Fig. 3.2.1 (1). Points Corresponding to Various Scores in a G a m e of Chance

its position arriving alternatively at M3, ~r3 or 3/4. The probabilities of all
possible transitions are entered in Table 3.2.1 (1) which determines the transitional
matrix (call it ~z). In two cases the transitional probabilities equal unity so that
the corresponding points remain where they are. These cases are characteristic
of the end of the game.
The maximal number of sets yet to be played is two. Therefore, the prob-
abilities of transitions [M 1-+(M3 or 3~r3)] and (Mt--+M4)as shown in matrix ~z2

(00
are ¼ and ¼ correspondingly:
0 0 ~ -
71,2 = 0
0 1 "
0 0 0
Of course this interpretation does not mean that PASCALreally introduced random
processes.1
It is also possible to interpret the problem discussed in terms of conditional
probabilities. Denote the event of A's winning the third set in PASCAL'Sexample

Table 3.2.1 (1)

Probabilities of transition from score to score; otherwise (see Fig. 3.2.1 (1)) from point to point

Initial positions New positions

M1 M2 M3 or ~r 3 M4
M~ o ~ 21 o
I
M2 0 0 ½
M3 or ~r 3 0 0 1 0
M+ 0 0 0 1

1 A similar reservation holds as regards quantities

f(k,r,p)=C~+ k xprq k, k = 0 , 1, 2 ....
(0 < p < 1, q = 1 - p and r is a natural number) which are usually called PASCAL'S distribution ([28],
§ 8 of chap. 6).
234 O . B . SHEYNIN

(problem (3; 2 : 1)) by A~, the contrary event by A 2 and, finally, the event of A's
winning the game as a whole by A. Then

P(A1) = P(A2) =½, P(A/AI) = 1, P(A/A2) =½,

2
P(A) = ~ P(A/AI)P(Ai) =3.
i=1

3.2.2. Binomial coefficients. PASCAL'S letter also contains a rule for dividing
the stakes and a remark concerning a game of dice. He considers the problem of
points for games of the type (n; a : 0), a = 1, 2,..., n - 1. Although he does not say
so, this is sufficient because games (n; a:b) with a > b are equivalent to games
((n - b); ( a - b) :0). Thus his is a general method.
PASCAL begins with a game (5, 1 • 0) noticing that the valeur of the first set
(A's expected gain due him after the first set over and above his own stake) is
equal to
½C 4 35
7 8
- (3.2.2.1)
½ + + + cs + 128

Indeed, the maximum number of remaining sets is eight, so that the value of the
first set (in terms of both staken) is

1 35
28 2 28 256

(the second term in the left-hand side is the prior probability of A's winning the
game).
For case (n; 1:0) the value of the first set in PASCAL'S sense is

C~222 (3.2.2.2)
Px- 22n- 2 •

With no explanation PASCAL included a table of values of each set for games
(n; 1:0) and n = 1, 2 .... , 5, 6. In particular, the value of the second set (again in
PASCAL'S sense) is
C~"--23 (3.2.2.3)
P2 = 22n- 3 •

Formulae (3.2.2.2) and (3.2.2.3) are not given by PASCAL though he did notice
that
1 . 3 . 5 ... (2n - 3)
P 1 - 2.4.6 ... ( 2 n - 2 )
and that P2 = P1-
Thus, in calculating values of various sets (probabilities), PASCAL used sums
of binomial coefficients. However, in neither of his letters does PASCAL refer to
the arithmetic triangle. PASCAL'S treatise on this triangle [74] published posthu-
mously is mentioned by FERMAT in his letter of 29 Aug. 1654 ([29], pp. 307-310).
 Early History of Probability 235

Possibly PASCALhad sent him a copy of this treatise (still unpublished or at least
not yet normally published).
PASCAL did not remark that sums of binomial coefficients could be used
irrespective of the score although he did just this in his treatise (see § 3.6).

3.2.3. Small differences between probabilities. As to PASCAL'S comment on

games of dice, he refers to an error committed by DE M~R~ who supposed that the
probability of scoring six points in four throws of a die should be equal to the
probability of scoring twelve points in 24 throws of two dice.
The probability of the first event is

P1 = 1 -- ~0.5177

(DE M~Rr~ calculated the ratio P1 : (1 - Pt)) and, of the second event,
( 3 5 ~ 24
P2 = 1 - t--f~-/ ~0.4913. (3.2.3.1)

DE M~m~ knew, obviously from experience, that P2 < 1. L'Arithm~tique se dkmen-

toit was his comment!
However, noticing the minuteness of the difference between PI and P20RJ~
([73], pp. 411--412) and VAN DER WAERDEN [97] do not believe in explanations
based on actual experience. ORE supposes t h a t DE M ~ knew two formulae
for calculating Pz--the correct one (3.2.3.1) and its rather crude substitutel--and
that DE M~gffs comment was provoked by the difference between results furnished
by these formulae. 2
Can a minute difference between probabilities be detected empirically? It
seems that even professional gamblers would have been unable to detect P 2 -
(1-P2)=0.0174. However, apart from common, even professional gamblers,
there possibly was a kind of gamblers eum businessmen who preferred safe ways
of fortune-making to taking chances. One such way would be to participate in
many games in which the probability of winning was (even if slightly) higher
than ½.
Considering that marine insurance (§ 2.3.1) and life insurance (§ 2.3.2) were
partly connected with betting and also that men of substance purchased life
annuities on a number of young lives simultaneously (§ 2.3.3), why could bets not
have been made on the results of many games simultaneously? Obviously, the

1 Quite consistently, OR~ adds that formula (3.2.3.1) was generally known; otherwise, he argues,
it would have been mentioned by PASCAL. This argument is not really convincing; the formula in
question could have well appeared elementary and thus not deserving special discussions (at least
not in correspondence with FERMAT).
2 In GALILEO'S time gamblers detected a difference of probabilities equal to 0.0385. As both
DAVID ([19], p. 66) and MAISTROV([66],p. 30) assume that this difference is equal to ~ I shall adduce
the whole argument. The comparison is between the probabilities of scoring 10 or 11 points with
three dice on the one hand and 9 or 12 points on the other. In itself, the first outcome (call it A) has
probability P{A}=2~766 and the second outcome (B) has P{B}=z-~6 whence P{A}-P{B}=I-~g.
Nevertheless, disregarding all other possible outcomes, gamblers were able to compare P {A/A or B} =
27 with P{B/A or B} =~z whence the difference sought is ~ = 0 . 0 3 8 5 .
3g
236 o.B. SHEYNIN

probabilities of various outcomes in these games had to be estimated beforehand,

but this was probably accomplished by small "reconnaissance" bets.
In a few of his problems DE Molvva~ ([21], problem 70 and others) introduces
both gamblers and Standers by. Who were these bystanders, unnecessary from
the mathematical point of view? Late comers unable to find a place for them-
selves or those in strained circumstances afraid to lay the usual stake? Hardly
would DE MOlVRZ introduce such characters. Were they not after all (petty?)
businessmen laying bets on many games simultaneously? I have also found
Spectateurs in one of MONTMORT'Sproblems ([69], p. 169): Pierre, Paul et Jacques
jouent au Brelan...Deux des Spectateurs...disputent...
A second indirect fact in favour of my conjecture is that a small difference
of 0.0160 between two probabilities was detected, possibly by observation, in
the 18th century,

when the Play of the Royal Oak was in use ([21], p. iii). The Odds against any
particular Point of the Ball were One and Thirty to One, continues DE MOIVVd~.
This intitled the Adventurers, in ease they were winners, to have thirty two Stakes
returned... ; instead of which they (received) but Eight and Twenty...the Master
of the Ball maintained that they had no reason to complain; since he would undertake
that any particular point of the Ball should come up in Two and Twenty Throws;
of this he would offer to lay a Wager...

The probability of gain by the Master of the Ball in this bet was

Pz2 = 1 - (3~)22 = 0.5004

while for 21 throws
P21= 1 - ( ~31) 21 =0.4 8 44.

The Master did not undertake to reduce the number of throws to 21; he possibly
knew that P21 < ½.
Lastly, I must note that in discussing the problem of small differences between
probabilities KENDALL ([54], p. 29) concludes that...relative chances were all
reached on the basis of intuition or trial and error in the games played up to the
middle of the seventeenth century.

3.3. Pascal-Fermat, letter dated 24 August, 1654 ([29], pp. 300-307)

In this letter PASCALdiscusses FERMAT'Smethod (votre m~thode, qui prockde
par Ies combinaisons) of solving the problem of points and maintains, referring
to ROBERVAL,that it could not be applied in the general case of three or more
gamblers. (Actually, ROBERVAL objected to this method even in the case of
two gamblers, but PASCALnoticed that in this case his objection was unwarranted.)
F~RMAT'Smethod, as described by PASCAL,consists in separately enumerating
combinations that lead to wins by each gambler and dividing the stakes in the
ratio thus obtained. For problem (n; ( n - 2 ) : (n-3)) the maximal number of
sets is four, but the game can also end after two sets. Would it be correct then,
asked ROBERVAL, to enumerate combinations occurring in four sets? Even if
 Early History of Probability 237

the game ended in two sets, answered PASCAL, it could be supposed to continue
fictitiously for two sets more. He could have added that exactly this was his
own tacit assumption (see § 3.2.2). For four sets the 16 possible combinations
of wins are aaaa (i. e., all four sets won by A), aaab (three sets won by A and one
set by B),..., bbbb. In all, the letter a is contained no less than twice in 11 combi-
nations and letter b no less than thrice in 5. Therefore, the stakes should be
divided between A and B in the ratio of 9 .
Consider now the case of three gamblers. Using the same method, PASCAL
finds that in some instances the game is won by two gamblers. Therefore he re-
commends returning to his own mOthode gOn~rale. For game (n; ( n - 1) : ( n - 2 ) :
( n - 2 ) ) the stakes should be divided as 17 : 5 : 5, PASCAL adds without proof.
Consider various combinations occuring in two (not three) sets of this game.
Obviously, (1) A wins with probability ~, and (2) the score equalizes with probability
~. Thus the share of A is
5 2 1 17

etc. QED. However, the term m~thode g~nOrale gives rise to doubt: first, PASCAL
had not used this term before; second, when calculating the values of different
sets (§ 3.2.2) he used another method.
Here I must note a disappointing error on PASCAL'S part ([29], p. 301): if
a game of two gamblers continues through four sets the total number of possible
combinations is, he says, 42 . Actually, it is 24. The end result is the same, which
would not be the case in general.

3.4. Fermat-Pascal, letter dated 25 September, 1654 ([29], pp. 310-314)

FERMAT defends his combinatorial method. First, the case of two gamblers
winning simultaneously could be reconciled with c o m m o n sense if the priority
of winning is additionally considered. Second, the probabilities of wins by each
gambler could be separately calculated by the combinatorial method for various
durations of the game. Thus in PASCAL'S example (n; ( n - 1 ) : ( n - 2 ) : ( n - 2 ) ) , A
wins the whole game with probabilities ½, ~ and ~ for durations of one, two and
three sets correspondingly. The total probability of his winning is 17
In passing, F~RMAT noticed that the total number of combinations in n sets
(n = 2 or 3) is 3". Most likely he realized that the total number of combinations
for m gamblers and any natural n would be m".
For the case of three gamblers combinations could be enumerated in FERMAT'S
method by using expressions of the type

(a+b+c)" (3.4.1)
where n is now the maximal number of remaining sets. Thus, again for~PAscAL'S
example, n = 3 and combination abb will occur thrice. Now these three occurrences
are abb, bab, bba so that only two of them (the first ones) are favourable for A
etc.
Denote the number of sets to be won by A, B and C by X, Y and Z corres-
pondingly,
X+ Y+Z=n.
238 O.B. SHEYNIN

Then the use of expressions such as (3.4.1) resembles the use o f " t r i p l e " generating
functions
1
3-~- (a + b + c)"

of distribution for the triplet {X, Y, Z} (31-is the probability of each set being
won by each gambler).

3.5. Another problem posed by Pascal

In a letter to HUYGENS CARCAVI ([47], t. 1, pp. 492--494) describes the follow-
ing problem which PASCAL proposed to FERMAT. Throwing three dice, A under-
takes to score 11 points, while B, under the same condition, undertakes to score
14 points. Each success counts for one point and the game is to last until one
of the gamblers is 12 points ahead of his partner. Required is the ratio of chances
of both gamblers.
FERMAT gave a correct answer: the chances of A and B are in a ratio of 1156:1
approximately. Indeed, in any set these chances are 27 and 15 correspondingly
so that the required ratio is (~)12= (9)t2~1157.
PASCAL also gave the correct answer actually calculating 2712 and 1512. Thus
he did not bother to cancel three out of the initial ratio Ys.
2v
The problem described, soon to be solved also by HUYGENS ([47], t. 1, pp.
505--507), was the first in which a probability of a run of events was calculated.
Indeed, it seems to be the problem in which probabilities rather than expectations
were first sought. See also § 4.1.

3.6. Pascal's treatise [74]

I shall begin by quoting two passages (pp. 478 and 482).
(l) Pour entendre les r@les des parties, la premikre chose qu'il faut consid&er,
est que l'argent que les joueurs ont mis au jeu ne leur appartient plus, car ils en ont
quittd la propridtO ; mais ils ont refu en revanche le droit d'attendre ce que le hasard
peut leur en donner...1 le r@lement de ce qui doit leur appartenir doit ~tre tellement
proportionnd d~ ce qu'ils avoient droit d'espkrer de la fortune, que chacun d'eux
trouve entikrement @al de prendre ce qu'on lui assigne, ou de continuer l'aventure
du jeu." et cette juste distribution s'appelle le parti.
(2) ...il ne faut proprement avoir @ard qu'au nombre des parties qui restent
gagner gt l'un et dl l'autre, et non pas au nombre de celles qu'ils ont gagndes...
PASCAL goes on to mention three methods to faire des parties. One of them is
just a straightforward use of expectations. II y e n a deux autre, adds PASCAL,
l'une par le triangle arithmdtique, et l'autre par les combinaisons. It would have
been more correct to distinguish between the criterion (expectation) and the
methods of calculating probabilities needed to use this criterion.
Here in contrast with his correspondence (see § 3.2.2) PASCAL not only
estimates the value of various sets (pp. 493-498) but also uses his arithmetic
triangle (pp. 488-489) for calculating sums of binomial coefficients in the general
1 These lines were repeated by MONTMORT([69], p. 73) in a somewhat differentwording.
 Early History of Probability 239

case (n; a : b). If the tabular form of defining a set of numbers is recognized on
a par with the analytical form, then, for the binomial distribution with p = q,
PASCAL'S method is equivalent to the method of generating functions. For this
reason I am inclined to begin the prehistory of generating functions with PASCAL.
Notice, however, that the tabular form of defining binomial coefficients prevented
PASCAL from generalizing his method of dividing the stakes so as to include the
case of three gamblers. Such generalization would have led to the coincidence
of the methods of PASCAL and FERMAT (see end of § 3.4).

3.7. Aleae geometria

In 1654 PASCAL communicated a letter ([75], pp. 101--103) A la trks illustre
AcadOmie Parisienne de science (the forerunner of the official Academy) informing
it about his desire to write a number of tracts. One of them was to be devoted
to geometry of chance:
...joignant la rigueur des demonstrations de la science (matheseos is the Latin
original) ~ l'incertitude du hasard, et conciliant ces choses en apparence contraire,
elle peut, tirant son nom des deux, s'arroger d bon droit ce titre stup~fiant : La
Gkomdtrie du hasard.
PASCAL'S wish did not come true but its mere existence proves that he wanted
to explicate elements of the theory of probability. An interesting attempt to
guess the subject-matter of PASCAL'S proposed tract is due to R~NYI [86]. One
may agree with R~NYI about the subject-matter, but not as regards the year
(1654) which he proposes as the date of PASCAL'S work or, rather, PASCAL'S
unknown letters to FERMAT. In any case, in his letter to the Academy PASCAL
mentions only one problem, that of dividing the stakes; no other problem is
even hinted at.
Another shortcoming of R~NYI'S attempt is his lack of reference to ARISTOTLE,
who ([93], § 2.2) had connected randomness with nonfulfillment (or nonexistence)
of goal. ARISTOTLE'S point of view would have hindered the development of
probability had it not been tacitly rejected both by HUYGENS and J. BERNOULLI
(or at least not upheld by them). As to PASCAL, the reasoning on philosophical
problems of probability put into his mouth by R~NYI seems hollow just because
he does not mention the Philosopher.

4. Huygens
I shall discuss HUYGENS' main work in probability in § 4.1. His correspondence
and manuscripts which contain interesting achievements pertaining to probability
are described in § 4.2, while § 4.3 is devoted to the problem of moral certainty
as explicated by HUYGENS.1

4.1. Huygens' main work [48]

This work is prefaced by a letter written by HUYGENS to VAN SCHOOTEN
(pp. 57--58), in which HUYGENS prophetically remarks that the study of games
1 In part 1 of his Horologium oscillatorium [49] HUYGENSconsiders various errors of pendulum
clocks. Research of this kind pertains to the prehistory of design of experiments [89], [92].
240 O.B. SHEYNIN

of chance is not a s i m p l e j e u d'esprit and that i t j e t t e l e s f o n d e m e n t s d'une spdcula-

tion f o r t int&essante et profonde. However, continues HUYGENS, studying these
games quelques-uns de plus COlObres Math~maticiens de toute la France...ont...
cachO leurs mkthodes (more accurately, did not publish their work) . . . . il m'est
impossible d'affirmer que nous sommes partis d'un mOme premiere principe.
(Precisely this was the case). M a i s . . . j ' a i constatd en bien de c a s q u e rues solutions
ne diffkrent nullement des leurs.
The criterion t-IuYGENS used was the mathematical expectation of a r a n d o m
event; the essence of his treatise actually consisted in explicating this concept
and its use in studying games of chance.
The first three propositions of the treatise are devoted to the expected gain
in a game of chance for cases of two or three equally probable gains and for
a case of two unequally probable gains. Neither here nor in his correspondence
does HUYGENS introduce a special term for this concept. Thus in proposition iii
he only describes expectation:
Avoir p chances d'obtenir a et q d'obtenir b, les chances Otant Oquivalentes,
me vaut
pa+qb
P+q
In § 3.1 I have noticed that no term for expectation was introduced either by
PASCAL or FERMAT.
In the following six propositions HUYGENS discusses the problem of points
in cases of two or three gamblers. Obviously he did not know either the elegant
combinatorial method due to FERMAT (§ 3.3) or the method of binomial coefficients
due to PASCAL (§ 3.2.2) and appropriate for the case of two gamblers. W h a t
HUYGENS actually did was to solve the problem of points for a number of sets
of initial conditions by a direct calculation of expectations and his proposition
ix includes a corresponding table of results thus obtained.
The last five propositions are devoted to problems connected with a game
of dice. I shall describe two such problems.
(1) [4.1 (1)]. In how m a n y throws should a gambler undertake to score twelve
points with two dice.
(2) [4.1 (2)]. A undertakes to score 7 points with two dice, and B undertakes
to score 6 points. They throw the dice alternately with B beginning the game.
Required is the ratio of their chances?

Problem [4.1 ( 1 ) ] . The probability of success in one throw is 3%. If, however,
the gambler is entitled to two throws, the expectation of his success is
1 _--35 1 -__ 71
gg u -t- Sg • Eg u -- i~gg u
where a is the stake. In four throws the expectation is
71 ..1_ 1 2 2 5 . 71
l~ a . l~-~ 1-~ga etc. 2

A few simple problems of the same kind were first solved by HUYGENSin 1656, in his corres-
pondence ([47], t. 1, pp. 426-427), without ever explaining the method of solution.
J'attends avec impatience ce qu'en dira Monsieur Fermat, wrote HUYGENS,pendant quoy vous
me permettrez de tenir cachee la solution...
2 I-IUYGENSalso solved a similar problem and, for that matter, by a similar method, in a manu-
script dating to 1676 ([47], t. 14, pp. 156-163).
 Early History of Probability 241

Describing this solution REIERSOL [85] interprets it as the use of formula

E(E(J(/Y))= E X (4.1.1)
in which, as I understand it, the first E corresponds to taking the mean among
y's. I shall return to formula (4.1.1) in § 4.2.3. Meanwhile, it seems that REIERSOI~'S
interpretation is not really convincing and that it would be more natural to
interpret HUYGENS' solution in terms of conditional probabilities. Also, for the
modern reader, the most natural method of solving this problem seems to use
the generating function
t 35
+ 364
Problem [4.1 (2)]. Let the expected gain of A at B's turn to play be x and let
it be y at A's own turn. B wins each time he plays with probability P2 = ~ . There-
fore
x -- ( 1 - P 2 ) Y = ~31
Y,

y=~a+~x
where Pl = 1 is the probability of scoring seven points and a the stake. It follows
that x=gra
31 so that finally the ratio sought is gg.
31
I shall discuss similar problems in §§ 4.2.1 and 4.2.2.
The treatise ends by formulating five additional problems (discussed in
§§ 4.2.1 and 4.2.2). Here are these problems the first and third of which are due
to FERMAT (see § 4.2.1) and the last to PASCAL (see § 3.5).
(1) [4.1 (1 add)]. A undertakes to score 6 points in a throw of two dice while
B undertakes to score 7 points. A begins with one throw, then B has two throws
after which each gambler in turn has two throws. Required is the ratio of the
chances of these gamblers.
(2) [4.1 (2 add)]. There are 12 counters, 8 of them black and 4 white. A, B
and C playing in turn draw these counters one by one. The gambler who first
draws a white counter wins the game. Required is the ratio of chances of these
gamblers.
(3) [4.1 (3 add)]. A pack contains 40 cards. Four of them are drawn. Required
is the ratio of chances that cards of all suits are or are not thus drawn.
(4) [4.1 (4 add)]. A gambler draws 7 counters out of the 12 mentioned in prob-
lem (2). Required is the ratio of chances that among these 7 counters 3 would
or would not be white.
(5) [4.1 (5 add)]. A and B undertake to score 14 and 11 points correspondingly
in a throw of 3 dice. They have 12 counters each and the winner receives one
counter from his partner. Required is the ratio of chances that they be ruined)
Problems [4.1 (2 add)] and [4.1 (4 add)] are not formulated clearly enough;
see §§ 4.2.1 and 4.2.2.

4.2. Huygens' correspondence and manuscripts

HUYGENS repeatedly returned to probability in his correspondence; besides
this he left manuscripts on the same subject, possibly written in connection with
No ruin occurs in this problem as formulated by PASCAL(see § 3.5).
242 O.B. SHEYNIN

his correspondence. These manuscripts are now published as appendices to his

main treatise ([47], t. 14, pp. 92-179). 1
4.2.1. The Year 1656. In 1656 CARCAW([47], t. 1, pp. 418--419) sent F~FtMAT
HUYG~NS' solution of the problem of points. HUYGENSalso offered an additional
problem on the same subject or possibly the same problem with another set of
initial conditions (ibidem, p. 432). FERMATsolved the additional problem to the
complete satisfaction of HUYGENS (ibidem, p. 442):
J'ay veu par la solution que M. de Fermat d faite de mon Probleme qu'il a la
methode universelle pour trouver tout ce qui appartient gt cette matiere...
Then, in a letter to CARCAVI,FERMAT(ibidem, pp. 433-434) proposed (to HUYGENS?)
five problems complete with answers without indicating the method of their
solution. Problems No. 1 and 4 were published by HUYGENS in his treatise as
being his problems [4.1 (1 add)] and [4.1 (3 add)]. The remaining three problems
are:
(2) [4.2.1 (2)]. A who begins the game undertakes to score 6 points in 2 throws
of 2 dice while B undertakes to score 7 points in 3 throws of 2 dice. Required is
the ratio of chances of their winning.
(3) [4.2.1 (3)]. A and B draw in turn one card at a time from a pack of 52
cards until the first heart appears; he who draws it is the winner. Required is
the ratio of chances of winning for these gamblers.
(5) [4.2.1 (5)]. 12 cards are drawn out of a pack of 36 cards. Required is the
ratio of chances that 3 aces occur or do not occur among the cards drawn.
HUYGENS solved these problems as follows.
Problem [4.2.1 (1)]=[4.1 (1 add)]. Denote ([47], t. 1, pp. 442-443) the
probabilities of scoring 6 and 7 points by PI and P2 correspondingly (Pl 5
P2 =~), A's share by x and the successive throws by A1, B1, B2, A2, A3 .... Assum-
ing a unity stake, the expectation of A's gain in throw A2 would be
p~ +(1 - p O x .
The expectation of A in previous throws would be
B2: (1-p2) [ p l + ( 1 - p l ) x ] ,
BI: (l-p2) z [&+(1-pl)x],
A1 : Pl +(1 - P l ) (1 - p 2 ) z [191+(1 - p l ) x ] .
The last expectation is equal to x:
P l -t-(1 - - P l ) (1 - - p 2 ) 2 [/)1 +(1 - p l ) x ] =x

10,355 which coincides with the answer given by FERMAT.

and x = 22,631 '
i Appendix 1 dated 1656 and possibly written in connection with HUYGENS' correspondence
with CARCAWis devoted to the problem of points for the case of three gamblers. The other appendices
were written during 1665-1688, the dates having been supplied in most cases by KORTEWEG, the editor
of t. 14 of the ~.uvr. compl.
 Early History of Probability 243

Problem[4.2.1 (2)] (~idem, p. 444). Only the final answer is given:

PA _ 87,451
P~ 72,360"
FERMAT'S answer supplied with the problem itselfis
PA 72,360
PB 87,451
Actually (seethe solution ofproblem[4.1 (2add)]in §4.2.2)

x= Pl l-x= P2(1-pl)
Pl+Pz-PlP2' Pl+Pz-PlP2
With
335 91
Pl -- and P2 =
1296 216
P A _ 335 .{ 961 91 . ' ] = 7 2 , 3 6 0
PB1296 \1296216] 87,451
which coincides with the answer given by FERMAT.
Problem [4.2.1(3).] (ibidem). HUYGENS solved this problem by a method
tantamount to using conditional probabilites.
Problem [4.2.1 (4)] = [4.1 (3 add)] (ibidem). Only the answer is given:
p 1000
1-p 8139

Of course, this answer corresponds to

(Ca10)* 1000
P C4o - 9139

Problem [4.2.1 (5)] (ibidem, pp. 444 and 446-447). HUYGENSdid not attempt
to arrive at a final answer but he did indicate that the method of solution is the
same as the method used for problem [4.2.1 (3)].
For comprehensiveness I shall repeat that in his correspondence HUYGENS
solved PASCAL'Sproblem [4.1 (5 add)]; see § 3.5.

4.2.2. The Year 1665. In 1665 HUYGENS returned to probability in his cor-
respondence with HUDDE. At first ([47], t. 5, pp. 305-311) the correspondents
discussed three problems: [4.1 (2add)], [4.1 (4add)] and a problem pertaining to
the game of croix ou pile:
(3) [4.2.2(3)]. Playing in turn A and B toss a coin. If pile turns up a ducat is
staked; if croix, the game ends and the successful gambler receives all the stakes.
A begins. Required is the expected benefit of B.
244 O.B. SHEYNIN

Soon HUDDE ([47], t. 5, pp. 348--351) generalized the last problem thus:
(4) [4.2.2(4)]. A has one white and two black counters; B has a number of
counters of both these colours. Each gambler draws a counter out of his own stock
and replaces it. Drawing a black counter the gambler stakes a ducat; drawing a
white counter he receives all the stakes. A begins. Required is the ratio of white
and black counters in B's stock for a just game.

Problem [4.1 ( 2 a d d ) ] . Its solution is contained in a manuscript of 1665 ([47],

t. 14, pp. 96-10l; published as Appendix 2 to HtJVGEN'S main treatise). Denote
the expectations of A, B, C by x, y, z correspondingly. If the stake is a,

4a+8z
x - - -
12

Here ~ is the probability of A winning the game in his first attempt and ~ is the
probability of his finding himself in the position of C, i.e. the probability of
gaining z. Similarly
8 8
y=~-x, z = 12 y"

Suppose now that the probabilities of success of A, B and C are p~, P2 and P3
correspondingly and assume a = 1. Then HUYGENS' solution would seem to result
in
x = p l +(1 -P3) z ]
y =(1 - P l ) x I 4.2.2.1)
z = (1 - P 2 ) Y
Pl Pl (1 --Pl) Pl (1 --Pl) (1 --P2)
X-D' Y D , z- D '

D = ~ Pi - ~ P i P j + Y', PlP~Pk.
i i<j i<j<k

This solution, correct for Pl =P2 =P3, is wrong in the general case in which
x+y+z#: l. The error in system (4.2.2.1) is easily seen: y and z only indirectly
depend on the corresponding probabilities (P2 and P3). The correct form of the
system is
1-1o 3
x=pl + z,
P3

_ P_.! (1 - P O x,
Y-- Pl

Z = P3 (1 - P e ) Y ,
Pl
whence
Pz P2 (1 -p~) P3 (1 -P2) (1 -p~)
X=D ' Y= D , z= D
 Early History of Probability 245

I ought to add that even for unequal probabilities p , , P2 and P3 HUYGENS

always arrived at a correct answer; his method of treating this general case (see
his solution of problem [4.1(ladd)] in §4.2.1 is even elegant. However, the
solution just described seems more natural.
HtJDD~ ([47], t. 5, p. 307) solved this problem assuming that the counters are
drawn without replacement.
... il arrive, is his comment, au tirage des fOres d Hoorn et en Frise, lors de l'Olection
du Magistrat.
He arrived at
x:y:z=232:159:104.
The correct answer due to J. BERNOULLI([3], part 1, p. 65) is
x . ' y . ' z = 77 :53 : 35 (=231:159:105).
Problem [4.1 (4add)J. HUYGENS solved it in the same manuscript ([47], t. 14,
pp. 96-101). He actually used both conditional probabilities and the formula (in
standard notation)
P(B) = ~ P(Ai)P(B/Ai).
i=1

I shall reproduce the beginning of his solution. Exactly three white counters
out of seven could be drawn only if among the first six counters drawn there are
either three or two white ones. Suppose there were three; the probability of
success is then ~. But suppose there were two; the probability of success is ½. In
their turn, each of these suppositions involves its own probability etc.
Problem [4.2.2(4)]. Denote the number of A's white and black counters by a
and b and the corresponding numbers in B's possession by e and d. With no
explanation given HUYGENS([47], t. 5, pp. 391--395) offers the following formula x :
aZ c d ad 2
e2= - e d-~ ~-- (4.2.2.2)
b(a+b) b
In his turn, likewise without explanation, HUDDE (ibidem, p. 385) offers quite
another formula
ad
c =-- (4.2.2.3)
a+b
which corresponds to his own interpretation of the problem (see below).
An examination of this problem is contained in Appendices 3 - 5 to HUYGENS'
main treatise ([47], t. 14, pp. 102--150). The discussion seems rather cumbersome
and its details obscure; essentially it consists in determining the expected gains
of the gamblers in each drawing and in summing up these expectations.
Also cumbersome is a similar method used by HUDDE ([47], t. 5, pp. 463--471).
The following solution seems simpler and more natural. Denote the pro-
babilities of winning and losing each set by A and B by Pl, ql and P2, q2 cor-

HUYGENShas erroneously interchanged the notation for black and white counters. I have
corrected this mistake. Another mistake in formula (4.2.2.2) is contained in the French version of
HUYGENS'letter though not in the original vernacular (ibidem).
246 O.B. SHEYNIN

respondingly. Of course,
a b c d
Pl= a+b ' ql- a+b ' Pe= c + d ' q2= c + d
Suppose that beginning the game A draws a white counter.1 Then with probability
q2Pl A wins one ducat from his partner; also, with probability q2qlpl he wins
two ducats etc. Simultaneously, he loses one ducat with probability qa q2P2, two
ducats with probability q~ q~P2, etc. His expected gain is thus
pl q2 (1 + 2 ql q2 + 3 +...)- ql q2p2 (1 + 2 q2 + 3 +...)
which should be multiplied by p l , the probability of the main supposition.
The second possibility is that A begins the game by drawing a black counter.
In this case, allowing at once for its probability, A's expected gain is
P~ ql q2 (1 + 2 q~ qe + 3 q~ q~ +...) -P2 ql (1 + 2 q~ q2 + 3 q~ q~ +...).
Finally, the expectation sought is
[Pl q2 (P~ -- qlPz) + ql (P~ q2 --Pc)] (1 + 2 ql q2 + 3 ~ q2 + . . . )
(the second factor is of course finite). In a just game this expectation should be
zero. Therefore,
p2 q2 - P I P 2 ql q2 +Pl ql q2 - - qlP2 = 0. (4.2.2.4)

This equation in P2 admits of being somewhat simplified. More important,

it is equivalent to HUYOENS' formula (4.2.2.2). HUI)DE'S formula (4.2.2.3) also
follows from this equation if the first two terms on its left-hand side are disregarded.
With pl--*0 equation (4.2.2.4) leads to p 2 ~ O and with pl--,1 to p 2 ~ l . Other
values of p~ furnish two positive values of P2 exactly one of which is less than
unity. Thus

1 +Pl ql - V( 1 +Pl qa) z - 4P 2 ql

132 =
2pl ql
Problem [4.2.2 ( 3 ) ] was considered by FRENICLE DE BESSY in his letter to
HUYGENS ([47], t. 5, pp. 489--490). Without sufficiently explaining his method,
he arrived at a correct value for the expected gain of B. I shall offer my own,
possibly simpler, solution of this problem.
If A refuses to throw first, he should stake ½ ducat; if B in his turn refuses to
throw first he should stake ¼ ducat, etc. Refusing to play altogether, A is to stake
1 1 1 21
-2- + -8- + 32-2 + " " - 32 ducat.
Similarly, B is to stake
1 1 1 21
ducat.
4 t--~- + - ~ + . . . . 64
I HUDDE([47], t. 5, p. 406) thought that in this case the game concludes with nobody winning
or losing.
 Early History of Probability 247

The game now becomes just and each gambler is entitled to a gain of
.21 21 "~ 63
+ 6 4 - ) ' 2 = 128 ducat

so that the expected loss of A (and the expected gain of B) is

21 63 21
ducat.
32 128 128
HUYGENS ([47], t. 5, p. 352) held that problem [4.2.2(4)]
est d'une autre genre que toutes celles que ['on trouve dans mes traitOs imprimdsl...
Many other problems of the autre genre could be proposed, Mais l'utilitO n'en est
pas assez grande ...

In itself, this opinion is understandable. Equating the expected gains of both

gamblers so as to determine a probability is in principle tantamount to calculating
(and comparing) expectations from given probabilities. But then, this problem
and problem [4.2.2 (3)] were the first in which a generalized random walk, i.e. a
walk with positive probabilities of transitions of a point into non-adjacent points
([28], chap. 14, § 8) was actually considered.

4.2.3. The Year 1669 (mortality). In 1669, corresponding with his brother
LODEWIJK, (CHRISTIAAN)HUYGENS turned to stochastic problems in mortality.
Thus probability was first applied beyond the realm of games of chance.
The correspondence most possibly initiated by LODEWIJKwas occasioned by
the publication of GRAUNT'S mortality table (see §2.4.3). Becoming interested in
calculating the expected duration of life for people of various ages, LODEWIJK
([47], t. 6, p. 483) wrote to his brother):
Selon mon calcul vous vivres environ jusquTt l'aage 56 ans et demij (actually
CHRISTIAANlived 66 years plus). Et moijjusqu'd 55.
He obviously based himself on his own calculations (ibidem, pp. 515-518).
According to GRAUNT'S table, 36 men out of a hundred live an average three
years only, 24 men, eleven years, etc., so that the mean duration of life for a newly
born infant is
36.3+24.11+...
= 18.22 years. (4.2.3.1)
100
In a similar way LODEWIJKcalculated the mean duration of life for men aged 6,
16, 26, etc. years, these ages having been directly shown in GRAUNT'S table. Thus
for age six the first sum in the numerator of formula (4.2.3.1) should be omitted
and the denominator correspondingly decreased by 36, etc.
For CHRtSTIAAN,then forty, the mean duration of life would be
42+44+46+51 • 4+...
-- 57.1 years, (4.2.3.2)
13

i The plural form is unclear. Did HUYGENSrefer to his manuscripts?

248 O.B. SHEYNIN

not 56 ans et demij. Indeed, six persons died from age 36 to age 46. Dividing the
interval [36; 46] in five equal parts, I arrive at three points (42; 44; 46) to the right
of point 40 and it is these points (moments of death) that I have introduced in
formula (4.2.3.2). I was unable to find the date of LODEWIJK'S birth, and hence
unable to check the estimate of his own duration of life.
CHRISTIAAN(ibidem, pp. 524-525) warned that
il ne s'ensuit pas que les 18 ans et 2 mois (the mean duration of life) ... soit l'age de
chaque personne crede ou conceue ... and that (p. 528) il est beaucoup plus apparent
qu'il [l'enfant conceu] mourra devant ce terme.
Continuing the correspondence, CHRISTIAAN(ibidem, pp. 531-532) introduces
the probable duration of life though not the term itself: Combien il reste raison-
nablement a vivre for a man of a given age, he asks. He shows that this duration
could be determined by a graphic procedure; once more (ibidem, p. 537) he
explains the essence of the probable duration of life and indicates its possible use:
Ce sont donc deux choses differentes que l'esperance ou la valeur de l'aage f u t u r
d'une personne et l'aage auquel il y a egale apparence qu'il parviendra ou ne par-
viendra pas. L e premier est pour regler les rentes a vie, et l'autre pour les gageures. 1
The graphic procedure for determining the probable duration of life was based
on a graph ([47], t. 6, plate inserted between pp. 530 and 531) which HUYGENS
constructed drawing a continuous curve through empirical points given by
GRAUNT'S table. Actually this graph corresponded to a curve
y = 1 - F(x)

where F ( x ) is the cumulative distribution function although it corresponds to an

unusual interval of admissible probabilities [0; 100].
I shall now consider two problems which HUYGENS(ibidem, p. 528) formulated
and partly solved in this correspondence.
(1) [4.2.3 (1)]. What is the expected period of time during which neither spouse
would die? Or, continues HUYGENS,
si on m 'avoit promis 100 francs au bout de chasque an qu'ils vivront ensemble, pour
combien seroit il juste qu'on rachetast cette obligation ? 2

(2) [4.2.3 (2)]. What is the expected period of time during which
(a) Forty persons aged 46 would die out?
(b) Two persons aged 16 would both die?
HUYGENS supposes that problem [4.2.3 (1)] does not essentially differ from
problem [4.2.3(2b)]? Following is his solution of the last problem (ibidem, pp.
526-531). According to GRAUNT'S table each of the two men has

1 The conclusion [99] that HUYGENS recommended, to use the probable duration of life rather
than its mean duration, is thus erroneous. See § 2.3.2 for a connection between life insurance and
betting. I also notice that both CHRISTIAAN(pp. 524--526) and LOI)EWlJK(pp. 484--485) have previously
mentioned betting on lives of men.
2 Thus, he proposes to determine the value of a life annuity on two lives.
3 The more so as no data on the difference of mortality of men and women was then available.
 Early History of Probability 249

15 chances to live an average of five years more,

9 chances to live an average of fifteen years more etc.
Suppose, says HUVGENS, that they both draw tickets with the duration of
their lives written on them. If A, the one to die sooner, draws a ticket with five
years, then B, the last survivor, will live not less than five years. More precisely,
his 15 chances to live 5 years change into 7½ chances to live 5 years and another
7½ chances to live 8 years. Other chances which remain constant being considered,
B's duration of life turns out to be 20.8 years. 1
Now if A draws a ticket with 15 years, then B will live not less than 15 years
and his chances change into 19½ chances to live 15 years, 4½ chances to live 18
years, etc. Other chances which remain constant being considered, B's duration
of life becomes 24.3 years.
Considering thus all possible suppositions, HUYGENS notes that the first of
them is accomplished 15 times, the second, 9 times, etc. and calculates the final
duration of B's life. He also explicitly notes (p. 530) that the same method could
be applied to find the duration of A's life.
Thus HUYGENS used conditional expectations of life and it is this calculation
which should have been pointed out by REIERSOL (see §4.1). More importantly,
I-IuYGENS contemplated using an order statistic and was able to calculate its
expectation for a discrete distribution.
HUYGENS did not solve problem [4.2.3(2a)] and of course calculations for
40 men were simply impossible in those times. However, he could have well
assumed that the last survivor would live almost until age 86, i.e. almost until the
last possible age according to GRAUNT.
In connection with a problem similar to [4.2.3 (2a)] HUVGENS (p. 538) arrived
at an incorrect conclusion: assuming an equal probability of death during each
year from 46 to 56, he simultaneously believed that during the first years mortality
would be higher
a cause que le nombre des personnes (of the studied group) est plus grand alors
qu'apres que la mort en a ost~ quelques uns. 2

Possibly he assumed a constant ratio between the numbers of dead and still liv-
ing which is the case in stationary populations. Actually no such assumption is
valid for his problem: for a distribution continuous and uniform in some interval
[a; b] n order statistics would divide this interval into (n + 1) approximately equal
parts. Correspondingly, the annual number of deaths should remain approximat-
ely constant.

4.2.4. Work during 1676-1688. In 1676 and then again in 1679 and 1688
HUYGENS ([47], t. 14, pp. 156--179) returned to games of chance. In 1676 he solved
a problem similar to one described in his main treatise [48]; in 1679 he studied a

1 20.3 years, the value given by HUVGENS, is a mistake.

2 HUYGENS continues: Voicy une question assez jolie qui me paroit bien plus difficile...et que je
n'ay pas encore calcul~e, mais je vois le moien de le faire. Deux personnes de 16 ans chascun, combien
peuvent ils esperer de vivre ensemble...? Item dans quel temps seront ils mort tous deux? Ce sont en
effect 2 questions differentes, et ou il y a penser a ehacune.
This is problem [4.2.3(2b)] discussed above so that the passage remains obscure.
250 O.B. SHEYNIN

game called bassette; in 1688 he investigated another game later called WALDE-
GRAAV'S game ([96], p. 122).
Here is a description of the latter. A plays with B. The loser deposits a ducat
while the winner goes on to play with C, the last gambler, etc. The one who wins
twice in succession wins the game and gets all the stakes.
HUYGENS considered a few versions of this game. However, as noticed by
KORTEWEG, the editor oft. 14 of the ~uvres eomplktes, he committed errors. More
important, his solution is not sufficiently explained, and in this sense it is much
worse than a clear solution due to DE MoIvm~ ([20], pp. 237-243; [21], pp.
132-159). Of course, the difference is readily understood: a manuscript couM be
badly written.

4.2.5. Huygens' analytical methods. In the words of KORTEWEG ([47], t. 14,

p. 20)
il Otait facile d ses successeurs immOdiats (J. BERNOULLI?) ... de dipasser sur
plusieurs points importants l'ceuvre de Huygens, au moyen de l'application de
l'analyse combinatoire. Et il faut ajouter que ses pr~d&esseurs, Fermat et Pascal,
se servient de m~rne avec avantage (mais comme nous le savons a l'insu de Huygens)
de cette analyse pour la rdsolution de quelques problemes de jeu.
However, there is yet another reason why HUYGENS' solutions are more com-
plex than necessary, i.e. his use of expectations rather than probabilities.
Consider problem [4.2.2 (4)]. The expected gains in various drawings differ
while the probabilities of each outcome are constant throughout. Even if HUYGENS
did notice this (at the time or subsequently) he possibly had no desire to better
his solution; having published his treatise [48] and solved a few important
problems in mortality, he did not see any new applications of probability and lost
interest in it.* This of course is only a conjecture but in any case thefondements
d'une sp&ulation fort int~ressante et profonde (see §4.1) happened to have been
laid by J. BERNOULLI.
One more point: precisely because HUYGENS used variable expected gains
he could not avoid equations in finite differences. Commenting on HUYGENS'
solution of the problem mentioned above, KORTEWEG ([47], t. 14, p. 135) noticed
that
... toutes ces Oquations (used by HUYGENS) ... se rdduisent ~ des cas particuliers de
l'Oquation
Xm=m A __½Xm_1 - - ~1X m + 1
(A is just 1 ducat).
Thus the name of HUYGENS should be connected with the history of equations
in finite differences.

4.3. Moral certainty

The concept of moral certainty was introduced by DESCARTES and in the
Logique des Port-Royal (see §2.2). HUYGENS, for his part, introduced moral
1 HUYGENScould expect nothing new in the work of DE W1TT [100]. Still, it seems somewhat
curious that he did not mention this work in his correspondence.
 Early History of Probability 251

certainty into natural science. 1 He first applied moral certainty in one of his letters
dated 1673 ([47], t. 7, pp. 298-300):
... la cause de Ia pompe et du siphon est avec une trOs grande vraisemblance
attribude d la pesanteur de Fair e t a son ressort. Parce que cette action de la pesan-
teur de l'air se manifeste dans cent expOriences .... dans les choses de physique il n'y
a pas d'autres demonstration[s] que dans le dechiffrement d'une lettre. Ou ayant
fait des suppositions sur quelques legeres conjectures, si l'on trouve qu'elles se veri-
fient en suite, de sorte que suivant ces suppositions de lettres on trouve des paroles
bien suivies dans la lettre, on tient d'une certitude tres grande que les suppositions
sont vraies, quoy qu'il n'y ait pas autrement de demonstration, et qu'il ne soit pas
impossible qu'on n'en puisse y avoir d'autres plus veritabIes.
... en matiere de physique il n'y a pas de demonstrations certaines, et ... on ne
peut sfavoir les causes que par les effects en faisant des suppositions fondees sur
quelques experiOnces ou phenomenes connus, et essayant ensuite si d'autre effects
s'accordent avec ces mesmes suppositions . . . . d'autant plus qu'on trouvera de
phenomenes conformes a l'hypothese, d'autant plus vraisemblable la doit on tenir.
Se souvenant pourtant tousjours qu'on n'a point de demonstration de sa veritO, et
qu'il peut s'offrir tel autre phenornene qui estant incompatible avec nostre supposO
principe le detruise absolument. Cependant ce manque de demonstration dans les
choses de physique ne doit pas nous faire conclure que tout y est egalement incertain,
mais il faut avoir egard au degrO de vraisemblance qu'on y trouve selon le nombre
des experiences qui conspirent a nous confirmer dans ce que nous avons supposO .... en
examinant et pesant bien ce degrO de vraisemblance que l'on a trouvO dans quelque
chose, l'on peut en tirer grande utilit& parce qu'on prevoit par les choses connues les
effects qui raisonnablement doivent suivre, lorsqu'on appliquera certaines matieres
d'une maniere nouveIle, ou que l'on fera telle chose pour obtenir tel effect.
Similar assertions are contained in the preface to the Traite de la Iumiere [50] :
in physics
les Principes se verifient par les conclusions qu'on en tire ... S~avoir lors que les
choses, qu'on a demontrkes par ces Principes supposez, se raportent parfaitement
aux phenomenes que l'experience a fait remarquer ; sur tout quand iI y e n a grand
nombre, et encore principalement quand on se forme et prevoit des phenomenes
nouveaux, qui doivent suivre des hypotheses qu'on employe, et qu'on trouve qu'en
cela I'effet repond d nottre attente. Que si toutes ces preuves de la vraisemblance se
rencontrent dans ce que je me suis proposd de traiter, comme il me semble qu'elles
sont, ce doit estre une bien grande confirmation du succks de ma recherche ...
Only once, in a letter dated 1691, does HUYGENS ([47], t. 10, p. 739) mention
DESCARTES :

... je ne suis pas tout a fait pour le Criterium de des Cartes. Parce que clans la
geometric mesme on s'irnagine souvent de cornprendre tres clairement des ehoses
qui sont fausses, II a reste donc tous jours a scavoir si l'on a compris clairement et

1 Its introduction into probability is due to J. BERNOULLIwho proved that the relative frequency
of the occurrence of a r a n d o m event in the case he discussed is in a sense morally certain to coincide
with the corresponding probability.
252 O.B. SHEYNIN

distinctement, ce que est assez douteux clans des longues demonstrations. Et de la

naissent les paralogismes. Je serais donc plus pour les divers degrez de vraisemblance,
laquelle dans certaines rencontres est si grande que d'estre quelque fois comme
10 ll et plus contre un, 1 que le vray ou le f a u x d'une proposition, et qu'en de certaines
choses eela va comme dl l'infini,
Thus HUYGENS seems to dismiss probabilities of the order of p = ( 1 + 1011)-1
= l 0-11 He does not explain the origin of his estimate, and it is doubtful whether
he ever used this or any other estimate as a criterion, but I notice that, according
to BOREL ([9], p. 27), p = 10 -6 is insignificant on the human scale and p = 10 -15
is insignificant on the terrestrial scale.
In a small work of 1690 [51] HUYGENSmaintained (p. 541 ; see also the editor's
comment on p. 532) that human judgement is only more or less probable and that
the degree of certainty of judgements should be assessed by c o m m o n sense. He
repeated this idea in his Cosmotheoros ([52], p. 688), a writing in which he upheld
and even used the thesis about the plurality of inhabited worlds: 2
... il y a beaucoup de degr~s de vraisemblance dont les uns sont plus proches de la
vdritd que les autres," c'est surtout clans l'~valuation de ces degr~s qu'on doit faire
preuve de bon sens.
I shall add that, according to the prevalent concept of his time ([93], pp.
132-135 and 140), HUYGENS ([52], p. 700) did not believe in the chance origin of
the world:
En effet ... un sectateur de Ddmocrite, ou bien aussi de Descartes peut se faire
fort d'expliquer tout les phdnomOnes Terrestres que les phOnomOnes cdlestes de
maniOre dl n'avoir besoin que d'atomes et de leurs mouvements ... jamais de pareils
objets n'ont pu Otre Ie r~sultat du mouvement dOrdgld et fortuit de corpuscules,
puisque l'on constate que tout y est parfaitement accomodd d~ de certaines fins...

5. General Conclusions
Games of chance served as models for posing natural and properly formulated
stochastic problems the solution of which led to the development of elements of
probability.
Jurisprudence was a source for disseminating stochastic ideas, which is
especially true as regards the lone work of N. BERNOULLI [4] on the application
of probability to jurisprudence. A special feature of this work is that BERNOULLI
calculated the expectation of an order statistic for a continuous uniform distri-
bution.
Moral certitude introduced into probability by J. BERNOULLI was first dis-
cussed by DESCARTES, then in the Logique de Port-Royal [2] and at least the
authors of the Logique if not DESCARTES explicitly advocated its use in juris-
prudence.

1 HUYGENSactually wrote out the eleven zeros.

2 L'improbabilitg de la thdse que parmi tousles corps cdlestes un seul, la terre, serait habit~, lui
parait extrdmement grande. (Editor's note on p. 534 of the appropriate volume of the ~Euvr. compl.)
 Early Historyof Probability 253

Marine insurance existed in two main forms:

(1) Bottomry, i.e. loans or mortgages on vessels with repayment complete
with high interest pending their safe arrival.
(2) Bets on the safe arrival of vessels, often prohibited.
These forms were hardly conducive to the development of probability.
Life insurance (life annuities excepted) was obviously underdeveloped partly
because of numerous prohibitions. This underdevelopment caused a negative
feedback: estimation of risks was hardly possible and claims did not occur
regularly enough. Consequently, business could not sufficiently base itself on
stochastic considerations.
Life insurance (in particular, insurance against accidents) possibly developed
from bets on the life of man, a practice which could have been readily adopted
from the corresponding form of marine insurance; it was primarily such bets
that lawgivers prohibited time and time again.
Calculation of the value of life annuities was the goal of J. DE WITT [100].
Assuming a certain law of mortality, he determined the expected amount of
annuity due to annuitants and thus opened up a new field for applying probability.
Transactions connected with life annuities (tontines excluded) were never
considered illegal or undesirable and, to a certain extent, they did foster devel-
opment of probability both practically (vital statistics was accumulated) and
theoretically (assumptions on mortality had to be made). However, at least in
some instances, statistics and methods of actuarial calculations remained un-
available.
Demographic statistics, the most important branch of political arithmetic,
originated in the works of J. GRAUNTand W. PETTY. GRAUNT'Stable of mortality
proved to be the origin of further work due to HUYGENS and a stimulus for
HALLEY [44], whose own table of mortality paved the way for estimating po-
pulations by means of birth and death data and exerted considerable influence
on DE Mo~vl~.
Thus, becuse of their general importance and particular application to life
insurance, mortality problems became most prominent in demographic statistics.
HALLEY'S data were furnished by NEUMANN, a preacher and scholar who
corresponded with LEIBN~Z. NEUMANN'S quantitative reasonings on mortality
influenced both LEIBNIZ and HALLEY.
LErBNIZ left several papers on political arithmetic which contained important
if simple conclusions, but unfortunately they remained unpublished.
The concept of mathematical expectation of a random event (gain or loss) was
actually introduced, if not formally defined, by PASCAL and FERMAT. These
scholars used expectation as a criterion so as to solve a number of problems in
games of chance, notably the problem of points, a venerable problem dating back
to the 14th century and considered classical since it was connected with PASCAL
and FERMAT.
Both these scholars used combinatorial methods for calculating the prob-
abilities sought. Also PASCALactually used conditional probabilities, while his
use of the arithmetic triangle could be considered as a substitute for the use of a
generating function for the particular case of the binomial distribution with p = q.
254 O.B. SHEYNIN

The discussion of games of chance by most eminent scholars stirred up the

scientific community and directed the attention of HUYGENS. His treatise [48]
was the first published work on probability and later served as a basis for part 1
of J. BERNOULLI'S Ars conjectandi [3]. HUYGENS' prophetic opinion to the effect
that the study of games of chance lays the foundation of a profound theory pos-
sibly stimulated additional interest in probability.
Leaving the realm of games of chance, HUYGENS arrived at interesting achiev-
evements pertaining to demographic statistics soon after the publication of
GRAUNT'S table. Thus HUYGENS calculated the expectation of an order statistic
for a discrete empirical distribution, introduced the mean and probable durations
of life 1 and constructed and used a graph of a continuous function
y = 1 - F(x),
where F ( x ) is the cumulative distribution function. Regrettably all these achieve-
ments, which today we know from the correspondence of HUYGENS, remained
unknown in their own time.
Developing the concept of moral certainty (see above) HUYGENS maintained
that in natural science (physics) the only possible proofs are those of stochastic
nature and that they should be checked by the truth of conclusions drawn f r o m
them.

Addendum to ~2.2, 2.3.2, 2.3.3 and 2.4.4

The time which passed after this article was sent to the editor proved unusually
eventful. First, volume 3 of J. BERNOULLI'S W e r k e [101] arrived in Moscow.
Expecting this source for quite a while, I never expected it to contain reprints of
the works of N. BERNOULLI [4] and DE WITT [100] complete with relevant com-
mentaries as well as a n u m b e r of contributions on the history of probability. In
particular, KOHLI & VAN DER WAERDEN [107] have mentioned a lesser known
and possibly comprehensive contribution of D o PASQUIER[114] on the history
of tontines. I have now inserted two passages from this contribution in my §2.3.2.
Second, VAN BRAKEL, commenting on the work of DE WITT (see my §2.3.3) has
proved ([103], pp. 130-131, footnote) that, after all, this work had become
available in the next two or three generations after its first appearance. VAN
BRAKEL also mentioned some sources not known to me before, notably the
Mdmoires [113]. These contain articles on DE WITT,HUYGENS,HUDDE, STRUYCK
and KERSSEBOOM as well as texts of Certificats de rente viagdre dating back to
1228 and 1229; 2 see my §2.3.2: Third, volume 2 of Studies in history o f statistics,
etc. [117] has appeared. A m o n g other reprints it contains those of articles written
by SEAL [115] and LAZARSFELD [108] which I did not k n o w about. Agreeing with
GREENWOOD (see my §2.3.2), SEAL does not attach any great importance to
ULPIANUS' table. SEAL also refers to two general sources on the history of in-

1 The mean duration of life was introduced also by LODEWIJKHUYGENS.

2 VAN BRAKELcriticized one of my articles [93] noting that my account often remains in a raw
condition because I did not distinguish between the various concepts of probability. The distinction
he wants to see is not discussed in the national mathematical literature and I had always restricted
myself so as to avoid a terra incognita.
 Early History of Probability 255

surance [104], [118]. The one which I managed to see so far [104] is well worth
reading; however, I do not hold myself guilty of omitting any essential infor-
mation on the subject.
I shall now take up (1) the new commentary [106] on N. BERNOULLI'Sdis-
sertation (see my §2.2); (2) LAZARSEELD'Scomment on the history of statistics
(see my §2.4.4).

(1) KOHLI (p. 541) remarks that Der geistige Vater dieses [N. BERNOULLI'S]
Werkes ist eindeutig Jakob. Ganze Abschnitte sowohI aus dem Tagebuch [102]
als auch aus der Ars Conjectandi hat Niklaus w6rtlich iibernommen. An andern
Stellen wurden Fragestellungen und blofle Andeutungen Jakobs aufgegriffen und
weiterverarbeitet.

I shall add that the Tagebuch was not even meant for publication. KOHLI
(ibidem) also provides a translation of a passage from the Praefatio to N. BER-
NOULLI'S w o r k . JAKOB BERNOULLI, NIKLAUS testifies, ...hat mir...die Veranlas-
sung gegeben...den Gebrauch der Mutmafiungskunst in Fragen des Rechtes zu
wiihlen. . . . ich sehe daft mit Hilfe der Mutkunst viele iiufierst wichtige Fragen, die
fast tdtglich vor Gericht behandelt werden, entschieden werden k6nnen, besonders
solche, welche Leibrenten oder die Toterkliirung yon Verschollenen betreffen.

(2) LAZARSEELO(p. 219) studies the battle between political arithmetic and
the German university statistics ( = Staatswissenschaft): The battle was won, in
Germany as well as elsewhere, by the political arithmeticians. From the beginning
of the 19 th century onwards, they also monopolized the title of statisticians. What-
ever was left of the former activities of university statisticians was thereafter con-
sidered a part of political science.

But why did university statistics originate and develop in Germany? Posing
this question, LAZARSFELDcompares (p. 221) PETTY with CONRING: The English-
man, citizen of an empire, looked for causal relations between quantitative variables.
The German, subject of one of 300 small principalities . . . tried to derive systematically
the best set of categories by which a state could be characterized.
International law, LAZARSFELDcontinues (p. 223), started [in Germany] a few
miles from everyone's house or place of business. No wonder then that it was a
spirit of systematically cataloguing what existed, rather than the making of new
discoveries, that made for academic prestige.
The author then introduces LEIBNIZ as a junior colleague of CONRIYG and
erroneously maintains (p. 226, footnote 29) that political arithmetic is about the
only topic of contemporary knowledge on which Leibniz himself did not write.
For my part, searching for political arithmetic in LEIBNIZ' writings, I have
omitted noticing, or perhaps even had been afraid of remarking, on the Staats-
wissenschafiliche aspect of his studies [60]-[62]. Now I maintain that LEmNIZ
had been both a political arithmetician and a university statistician at the same
time but that even he failed to root political arithmetic in Germany.

Acknowledgement. Professor J. COHEN and Dr. W. ROMBERG have helped me to obtain necessary
literature.
256 O.B. SHEYNIN

References
1. ADLES, L., Note on the early history of life assurance. Assurance mag., vol. 3, 1853, p. 64.
2. ARNAULD,A., & P. NICOLE, Logique de Port-Royal (1662). Paris, 1877.
3. BERNOULLI,J., Ars conjeetandi (1713). German transl. : Leipzig, 1899.
4. BERNOULLI, N., Dissertatio inauguralis math.-juridica de usu artis conjectandi in jure. Basileae,
s.a. [1713].
5. BIEDERMANN, K., Von und aus noeh ungedruckten Leibniz'sehen Handschriften. Westermann's
Monatshefte Jg. 26, Bd. 52, Juli 1882, 453-462.
6. BIENAYM~,I. J., Sur un effet de l'intOr~t composk. ProcOs verb. Soc. philom. Paris, 1839, 60-65.
(Extrait de L'Institut, J. gOnkral sociOt~s et travaux scient., 1e sect., No. 286.)
7. BIERMANN, K.-R., Problems of the Genoese lottery in the works of elassics in probability. Hist.-
math. essays, vol. 10, 1957, 649-670 (in Russian).
8. BOCKH, R., E. Halley als Statistiker. Bull. Inst. intern, stat., t. 7, No. 1, 1893, 1-24.
9. BOREL, 1~., Lesprobabilitks et la vie (1943). Engl. transl. : New York, 1962.
10. CANTOR, M., Politische Arithmetik. Leipzig, 1898.
11. CANTOR, M., Vorlesungen iiber Geschiehte der Mathematik, Bd. 3. Leipzig, 1901.
12. CELLINI, B., Autobiography. New York, Modern library ser., s.a.
13. CI-L~UFTON,A., Les assuranees, t. 1. Paris, 1884.
14. COHEN, J., & E. I. CHESNXCK,The doctrine of psyehological chances. Br. J. Psychol., vol. 61, No. 3,
1970, 323-334.
15. COHEN, J., E. I. CHESNICK& D. HARAN, Evaluation of compound probabilities in sequential choice.
Nature, vol. 232, No. 5310, 1971, 414-416.
16. COMMELIN,C., Beschryvinge der Stadt Amsterdam, t. 2. Amsterdam, 1693.
17. CONDORCET, Discours sur l'astronomie et le calcul des probabilit~s. (Lu 1787). (Euvr. compl., t. 1.
Paris, 1847-1849, 482-503.
18. COUTURAT,L., La logique de Leibniz. Paris, 1901.
19. DAVID, F. N., Games, gods and gambling. London, 1962.
20. DE MOIVRE, A., De mensura sortis. Philos. trans. Roy. soe. Lond., vol. 27, 1711 (1712), 213-264.
21. DE MOIVRE,A., Doctrine of chances. London, 1756 (posthumous edition). Repr. : New York, 1967.
22. DE MOIVR~, A., Treatise of annuities on lives (1725). ([21], 261-328).
23. DESCARTES,R., Lesprineipes de laphilosophie (1644). (Euvr., t. 9, No. 2. Paris, 1971 (whole issue).
24. ELSNER, E., Entwicklungslinien der Statistik. Humanismus und Technik, Bd. 18, 1974, 132-155.
25. t~MERIGON,B.-M., Traitk des assurances etc., t. 1. Marseille, 1783.
26. ENESTR6M, G., Sur la mkthode de J. de Witt (1671) pour le ealcul de rentes viagkres. Arehief voor
de verzekerings-wetenschap, vol. 3, No. 1, 1897, 62-68. Originally in Swedish (1896).
27. FEDOROVITCrLL. V., History and theory of statistics. Odessa, 1894 (in Russian).
28. FELLER, W., Introduction to probability theory, etc., vol. 1. New York, 1957.
29. FERMAT,P., O~uvres, t. 2. Paris, 1894.
30. FOURIER,J. B. J., Rapport sur les tontines, 1821 (1826). (Euvres, t. 2. Paris, 1890, 617-633. Prepared
by POISSON, LACROIX, FOURIER-- rapporteur.
31. FREUDENTHAL,H., Probability and statistics. Amsterdam a.o., 1965.
32. FREUDENTHAL,H., & H. G. STEINER, Aus der Geschiehte der Wahrscheinliehkeitsreehnung undder
mathematisehen Statistik. In: Grundziige der Mathematik, Bd. 4. Hrsg., H. BEHNKE et al.
G6ttingen, 1966, 149-195.
33. GINh C., Gedanken zum Theorem yon Bernoulli. Sehweiz. Z. fiir Volkswirtschaft und Statistik,
82. Jg., No. 5, 1946, 401-413.
34. GLASS, D. V., J. Graunt and his Natural and political observations. Proe. Roy. soc. Lond., vol.
B159, No. 974, 1963, 2-37.
35. GRAETZER,J., E. Halley und C. Neumann. Breslau, 1883.
36. GRAUNT, J., Natural and political observations made upon the bills of mortality (1662). Ed., W. F.
WILLCOX. Baltimore, 1939.
37. GREENWOOD, M., Graunt andPetty. J. Roy. stat. soc., vol. 91, pt. 1, 1928, 79-85.
38. GREENWOOD, M., A statistical Mare's nest? Ibidem, vol. 103, pt. 2, 1940, 246-248.
39. GREENWOOD,M., Medical statistics from Graunt to Farr. (Biometrika, 1941-1943). Repr. : Studies
in History of probability and statistics. Eds, E. S. PEARSON& i'¢J[.G. KENDALL. London, 1970,
47--120.
 Early History of Probability 257

40. GUHRAUER,G. E., Leben und Verdienste Caspar Neumann's. Nebst seinem ungedruckten Brief-
wechsel mit Leibniz. Sehlesische Provinzialbldtter, N. F. Bd. 2, 1863, 7-17, 141-151,202-210,
263-272.
41. GuY, W. A., Statistical development, with special reference to statistics as a science. Jubilee volume.
Roy. Stat. Soc. Lond., 1885, 72-86.
42. HAAS, K., Die mathematischen Arbeiten yon J.H. Hudde (1628-1704), etc. Centaurus, Bd. 4,
No. 3, 1956, 235 284.
43. HACKING,I., Emergence of probability. Cambridge, 1975.
44. HALLEY,E., An estimate of the degree of mortality of mankind, ete. 1692-1693 (1694). Baltimore,
1942.
44a.HALLEY, E., Correspondence andpapers. Ed., E. F. MACPIKE. Oxford, 1932.
45. HENDRIKS,F., Contributions to the history of insurance, etc. with restoration of ...De Witt's Treatise
on life annuities. Assurance mag., vol. 2, 1852, 121-150, 222-258; vol. 3, 1853, 93-120.
46. I-IENDRIKS,F., Notes on the early history of tontines. J. Inst. actuaries, or Assurance mag., vol. 10,
1863, 205-219.
47. HUYGENS,C., (Euvres eompl~tes, tt. 1-22. La Haye, 1888-1950. Editors, as named in the Avant-
propos in t. 22, of tt. 1 and 5--D. BIERENSDEHAAN;of t. 6, J. BOSSCHA;of t. 14, D. J. KORTEWEG.
48. HUYGENS,C., De calcul dans lesjeux de hasard (1657). ([47], t. 14, 49-91).
49. HUYGENS,C., Horologium oscillatorium, etc. (1673). ([47], t. 18, 27 438).
50. HUYGENS,C., Traitd de la lumikre (1690). ([47], t. 19, 450 548).
51. HUYGENS,C., R~flexions sur la probabilitO de nos conclusions etc. (1690). ([47], t. 21, 529-568).
52. HUYGENS,C., Cosmotheoros (1698). (Ibidem, 653-842).
53. JOHN, V., Geschichte der Statistik. Stuttgart, 1884.
54. KENDALL, M. G., The beginning of a probability calculus. (Biometrika, 1956). Repr. : Studies Hist.
Prob. Stat. (see ref. [39]), 19-34.
55. KENDALL,M. G., Where shall the history of statistics begin? (Biometrika, 1960). Repr. ibidem,
45-46.
56. KRUSKAL,W. H., & R. S. PIETERS,T. Paine and social security. In: Statistics by example. Exploring
data. Addison-Wesley Innovative ser. Reading (Mass.) a.o., 1973, 105-111.
57. LEIBNIZ, G. W., Demonstrationum Catholicarum conspectus (1668 16697). Siimtliche Schriften
und Briefe, 6. Reihe, Bd. 1. Berlin, 1971, 494-500.
58. LEIBNIZ,G. W., Elementa juris naturalis (1669-16707). Ibidem, 431-485.
59. LEIBNIZ, G. W., Die Werke gemglfl seinem handschriftlichen Nachlasse in der Kgl. Bibliothek zu
Hannover. Hrsg. O. KLOPP. 1. Reihe, Bd. 5. Hannover, 1866.
60. LEIBNIZ,G. W., Entwurfgewisser Staatstafeln. ([59], 303 314).
61. LEIBNIZ,G. W., Von Bestellung eines Registratur-Amtes. ([59], 315 320).
62. LEIBNIZ,G. W., Vorschlag zu einer Medizinal-Beh6rde. ([59], 320 326).
63. LEIBNIZ,G. W., Essay de quelques raisonnemens nouveaux sur la vie humaine et sur le nombre des
hommes. ([59], 326-337).
64. LEIBNIZ,G. W., Quaestiones calculi politici circa hominum vitam, et cognatae. ([59], 337-340).
65. LEIBNIZ,G. W., Siimtliehe Schriften und Briefe, 1. Reihe, Bd. 8, Berlin, 1970.
66. MAISTROV,L. E., Probability theory. A historical sketch. New York and London, 1974. (Orig. in
Russian, 1967).
67. MARMONIER,H., Absence. La Grande EncyclopOdie, t. 1. S.a., 145-147. According to Grand
Larousse, t. 4, 1961, p. 527, this Encyclopddie was published during 1885-1892.
68. MEITZEN,A., Geschichte, Theorie, und Technik der Statistik. Stuttgart und Berlin, 1903.
69. MONTMORT,P. R., Essay d'analyse sur lesjeux de hasard. Paris, 1713 (2nd edition).
70. MRO~EK, V. R., Entstehung und Entwicklung der Wahrscheinlichkeitsrechnung. Proc. Inst. hist.
sci. and technology, ser. 1, No. 2, 1934, 45-60 (in Russian).
71. NoGucm, S., Die Entwieklung des Versicherungsgedankens" in Japan. Z. fiir die ges. Versieherungs-
Wiss., Bd. 25, No. 3, 1925, 238-253.
72. O'DONNELL,TERENCE,History of life insurance. Chicago, 1936.
73. ORE, O., Pascal and the invention of probability theory. Amer. math. monthly, vol. 67, No. 5, 1960,
409-419.
74. PASCAL,B., TraitO du triangle arithmOtique (1665). (~uvres, t. 3. Paris, 1923, 445-503.
75. PASCAL,B., (~uvres completes. Paris, 1963. (One-volume ed.).
76. PETTY,W., Eeonomic writings, vols 1-2. Ed., C. H. HULL. Cambridge, 1899.
 258 O.B. SHEYNIN

77. PETTY,W., A treatise of taxes and contributions (1662). ([76], vol. 1, 1-97).
78. PETTY, W., Discourse read before Royal society. London, 1674.
79. PETTY, W., Political arithmetic (1690). ([76], vol. 2, 239-313).
80. PETTY, W., Verbum sapienti (1691). ([76], vol. 1, 99-120).
81. PETTY, W., Papers, vols 1--2. London a.o., 1927.
82. Petty-Southwell correspondence 1676-1687. London, 1928.
83. PTOUKHA,M. J., Graunt, fondateur de la d~mographie (1620-1674). In: Congr. intern, de la popu-
lation, Paris, 1937. 2. DOmographie historique. Paris, t. 2, 1938, 61-74.
84. REICnER, V. K., Socio-historical types of insurance. Moscow-Leningrad, 1947 (in Russian).
85. REIERS0L, O., Notes on some propositions of Huygens in the calculus of probability. Nord matem.
tidskr., t. 16, No. 3, 1968, 88-91.
86. R~NYI, A., Briefe fiber die Wahrseheinliehkeitsrechnung. Budapest, 1969. (Orig. in Hungarian.)
87. ROMEIN, J., & ANNIE ROMEIN-VERSSCHOOR, J. de Witt. In collection of authors' essays Anherren
der H61landischen Kultur. Bern, 1946, 277-312. (Orig. in Dutch.)
88. SEAL, H. L., Mortality data and the binomial probability law. Skand. aktuarietidskr. No. 3-4,
1949, 188-216.
89. SHEYNIN, O. B., Newton and the classical theory of probability. Arch. hist. ex. sci., vol. 7, No. 3,
1971, 217-243.
90. SHEYNIN, O. B., Finite random sums (a historical essay). Ibidem, vol. 9, No. 4-5, 1973, 275-305.
91. SHEYNIN, O. B., R. J. Boscovieh's work in probability. Ibidem, 306-324.
92. SttEYNIN, O. B., Mathematical treatment of astronomical observations (a historical essay). Ibidem,
vol. 11, No. 2-3, 1973;97-126.
93. SHEYNIN, O. B., Prehistory of the theory of probability. Ibidem, vol. 12, No. 2, 1974, 97-141.
94. STRUYCK,N., Hypothkses sur l'Otat de l'espece humaine (1739 or later). CEuvres. Amsterdam, 1912,
165-249.
95. STRUYCK, N., D~couvertes plus dOtaill~es concertant l'dtat du genre humain etc. (1752 or later).
Ibidem, 250-423.
96. TODHUNTER, L, History of the mathematical theory of probability (1865). New York, 1965.
97. VAN DER WAERDEN, B. L., Die Korrespondenz zwischen Pascal und Fermat iiber Wahrscheinlich-
keitsprobleme. Hist.-math. essays vol. 21, 1976, 228-232 (in Russian).
98. WALKER, HELEN M., A. De Moivre (1934). ([21], 351-368).
99. WRITE, C., & R. J. HARDY, Huygens 'graph ofGraunt "sdata. Isis, vol. 61, No. 1 (206), 1970, 107-108.
100. DE WITT, J., Waerdye van lyf-renten naer proportie van los-renten. I n ' s Graven-Hage. By JACOBUS
SCHELTUS, 1671. Author's name appears at end of contribution. A few lines in which HUDDE
declares his agreement with the principles of DE WITT'S calculations are inserted in the text.
Engl. transl. : Value of life annuities in proportion to redeemable annuities ([45], 232-249).
101. BERNOULLI,J., Werke, Bd. 3. Basle, 1975.
102. BERNOULLI,J., Aus der Meditationes (= Tagebuch ) ([101]. 21-89).
103. VAN BRAKEL, J., Some remarks on the prehistory of the concept of statistical probability. Arch.
Hist. Ex. Sci., vol. 16, No. 2, 1976, 119-136.
104. BRAUN, H., Geschichte der Lebensversicherung und der Lebensversicherungstechnik. Berlin, 1963
(2.Auflage). First pub1. 1925.
105. CONDORCET,Absent (1789). In: CONDORCET, Math~matique et soci~tO. Paris, 1974, 8-11.
106. KOHLI, K., Kommentar zur Dissertation yon N. Bernoulli ([101], 541-556).
107. KOI-ILI,K.,& B. L. VAN DER WAERDEN~Bewertung yon Leibrenten ([101], 515-539).
108. LAZARSFELD,P. F., Notes on the history of quantification in sociology-trends, sources and problems.
(Isis, vol. 52, 1961) ([117], 213-269).
109. LEIBNIZ,G. W., Allgemeine Untersuchungen iiber die Analyse der Begriffe und wahren Sdtze (1686,
ms) ([111], 241-303).
110. LEmNIZ, G. W., Assecuranzen. Werke, Bd. 6. Hannover, 1872, 231-242.
111. LEIBNIZ, G. W., Fragmente zur Logik. Berlin, 1960.
112. LEmNIZ, G. W., Philosophische Schriften, Bd. 3. Berlin, 1887.
113. Mdmoires pour servir dt l'histoire des assurances sur la vie et des rentes viag&es aux Pays-Bas.
Amsterdam, 1898.
114. Du PASQUIER,L. G., Die Entwicklung der Tontinen bis auf die Gegenwart ; Geschichte und Theorie.
Z. schweiz. Stat., J. stat. Suisse, 46. Jg., No. 5, 1910, 484-513.
115. SEAL,H. L., A budget of paradoxes. (J. Inst. Actuaries Students" Soc., vol. 13, 1954) ([117], 24-29).
 Early History of Probability 259

116. SOFONEA,T., Leibniz und sein Project zur Errichtung staatlicher Versicherungsanstalten. Schwe&.
Versicherungs Z., Bd. 65, 1957, 144-149.
117. Studies in the history of statistics and probability, vol. 2. Eds. Sir MAURICE KENDALL & R. L.
PLACKETT. London, 1977.
118. TRENERY, C. F., The origin and early history of insurance, etc. London, 1926.

Mishin Street 12, Flat 35

Moscow 125083

(Received November 24, 1976)