Mathematical Association of America

Definitions of Probability
Author(s): Edward S. Allen
Source: Mathematics News Letter, Vol. 8, No. 6 (Mar., 1934), pp. 132-135
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/3027639
Accessed: 21-10-2015 21:02 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact [email protected].

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to Mathematics News
Letter.

http://www.jstor.org

This content downloaded from 134.184.26.108 on Wed, 21 Oct 2015 21:02:37 UTC
All use subject to JSTOR Terms and Conditions
 Definitions of Probability

By Edward S. Allen
Iowa State College

Of the various definitions on which I wish to report, the most

widely accepted, of recent years, is the frequency definition. According
to it, the probability that, if O occur, P will also happen, is the limit,
as the total number of cases considered becomes infinite, of the ratio
of the number of cases in which we have O and P to the number in
which we have O. There is, I think, no harm in including the possibility
that there are but a finite number of terms in the series. It will be
noticed that P can have no probability in itself, and that probability,
as here defined, always refers to a particular ordered sequence. There
is?to illustrate the first point?^no sense in saying, "the probability that
a good die fall with the 3 uppermost is 1/6" We must precede that
with the statement, "if the die is thrown." Should we change the
conditional clause so as to read, "if the die has 3 uppermost when
1 centimeter above the table," there may well be a probability, but
it will not be 1/6.

According to the definition just given, a probability is that of

an event. Keynes, on the other hand, is emphatic in saying that only
propositions have probability. A bridge between these two points of
view was offered by Ancillon as early as 1794, when he said, "To
state that a fact is probable is to state that a proposition is probable"?
the probability that an event will happen is the probability that the
prediction of it will be fulfilled.
A more important divergence of Keynes* theory from the fre?
quency theory with which we started is his psychological point of view,
which leads to an interesting kind of topological probability. The psy?
chological point of view was not, of course, new with Keynes. Laplace
defined probability as "belief in the truth of a statement." To illustrate
this definition, let us consider the probabilities that (a) it rain tomorrow,
(b) it rain tomorrow and then snow, (c) it rain tomorrow and then
clear, (d) that a random Fourier series converge to the function from
which it was derived. Keynes would, I presume, refuse to assign a
numerical probability to any of these events; nevertheless, he would
have to give (a) a higher rank than (b) or (c), while holding himself
entitled to refuse to compare any other pair named.

This content downloaded from 134.184.26.108 on Wed, 21 Oct 2015 21:02:37 UTC
All use subject to JSTOR Terms and Conditions
 MATHEMATICS NEWS LETTER 133

Ramsey, who goes even farther than Keynes in his psychological

bias, proposes a method for measuring the degree of belief which
he identifies with probability. Let us see how he defines a probability
of l/2> by a combination of beliefs and preferences. We will suppose
that you definitely prefer sauerkraut to cheese. You are told, "Either
you are to have cheese if a coin, when tossed, comes up heads and
kraut if it comes up tails, or the other way about, Which shall it be?"
If you then say, "I don't care", that proves that, in your opinion, the
probability of appearance of heads is l/2. But, after all, does not this
indifference signify a belief that, in the long run, heads and tails
will appear equally often?
The classical definition of probability was given, perhaps for
the first time, by Pascal, It is the "ratio of the number of favorable
cases to the number of equally possible (or probable) cases." A funda?
mental difficulty with this theory is the need for a theory of probability
for the definition of "equally possible." It has been proposed that
complete ignorance be the basis for "equal possibility"; it has been
said that we should have a considerable body of knowledge, but some
ignorance, But, after all, this is a rather vague foundation for such a
precise number as .5; it seems to me better to return to the first
definition given in this account, and to say that two events are
equally possible if the ratio of the numbers of their occurrences
approaches 1.
Of late years, the vogue of axiomatization has reached the theory
of probability, Among the most important of the systems built up
from independent postulates are those of Reichenbach and of Tornier.
The former associates with two occurrences O and P a number called
the probability from O to P:? a member which, however, is given
no definition in terms of the objective world, The axioms are those
of uniqueness, normalization, addition and multiplication, The axiom
of normalization, for example, assigns to certain implication the num?
ber 1, and debars negative probability, That of multiplication,?"the
product of the probabilities from O to P and from O and P to Q is
the probability from O to P and Q." These suffice as a foundation
for the whole calculus.
Tornier's theory depends on the topology of null spaces. For him
a point is a sequence of elements of unspecified nature. The space
of such points has metric character; for the distance between two
points can be satisfactorily defined as the reciprocal of the number
of the first place at which the sequences differ. Certain point-sets in

This content downloaded from 134.184.26.108 on Wed, 21 Oct 2015 21:02:37 UTC
All use subject to JSTOR Terms and Conditions
134 MATHEMATICS NEWS LETTER

this space?in particular, the sets of points whose sequences agree up

to a certain place?can be assigned measures and,?with appropriate
normalization?the measures can be called probabilities. In more
familiar language, we have the probability that a given sequence of
events be followed by certain others.

Noteworthy in both these theories is their caution. Reichenbach

does not state that a probability is associated with every pair of
occurrences (it would be better to say, types of occurrences); Tornier
does not give assurance that every set in his nulLspace has a measure.
Both sets of axioms, on the other hand, are consistent with the
concept of probability as the limit of relative frequency,?wherever
that limit exists.

At this point, however, comes a doubt as to the exact applicability

of any theory to nature. We can never know that the fraction of
throws of a die resulting in a 3 approaches the limit 1/6?or any
limit whatever. What is, then, the validity of any calculus of
probabilities? Mathematically, its deductions can be correct. Physically,
it can be justified but partially, just as any statement as to the outer
world is edged with uncertainty.

Is it conceivable that a new definition of physical probability could

lay better claim to correspondence with reality, and at the same time
fit into our mathematical theory? Were we to say that a
probability
is not a limit of frequency within a sequence of events, but the sequence
itself, a mathematical theory could be built up, quite equivalent to
the current ones?in return for its greater claim to certain truth, it
would have lost all usefulness.

It has been suggested that a probability might be, not a number,

but an interval. The mechanical difficulties of a
corresponding theory
would be much greater. What is worse is the retention of
conceptual
difficulties. When we say "an interval," we mean, of course, an
interval with welLdefined ends; and a doubt as to the existence of a
frequency-limit leads inevitably to doubt as to the physical signifi-
cance of any definite numbers?other than 0 and 1.

In sum, all theories which have been constructed, if

they have
physical meaning at all, can be interpreted as frequency theories. The
inadequacy of even this interpretation can not be overcome; but the

This content downloaded from 134.184.26.108 on Wed, 21 Oct 2015 21:02:37 UTC
All use subject to JSTOR Terms and Conditions
 MATHEMATICS NEWS LETTER 135

practical benefits coming from its use justify taking it as a working

hypothesis.

REFERENCES

Ancillon?Doutes sur les bases du calcul des probabilities, Mem. Ac. Berlin,
1794-5.
Laplace?Essai philosophique des probabilites, Paris, 1814.
Keynes?A tveatise on probability,London, 1921.
Ramsey? The [oundations o[ mathematics,and other logical essays, London and
New York, 1931.
Reichenbach?Axiomatik der Wahrscheinlichkeitsrechnung Mathematische Zeits
chrift,vol. 34, 1932.
Tomier?Grundlagen der WahrscheinlichkeitsrechnungActa Mathematica, vol.
60, 1933.

Magic Squares

By H. Vann Parker

This note is concerned with a certain type of magic square which

I have found to be quite interesting, and which I hope may prove of
interest to readers of the News Letter. Such squares and the facts
concerning them are perhaps well known, but I do not recall having seen
a discussion of them. I shall give two examples and discuss some of the
characteristics of the squares.

A magic square with 25 numbers is

This content downloaded from 134.184.26.108 on Wed, 21 Oct 2015 21:02:37 UTC
All use subject to JSTOR Terms and Conditions