Scribd
Upload
17 views1 page

Hacking, Ian - Statistical and Inductive Probabilities (Review) (1964)

Uploaded by

themistoclesxii
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
17 views1 page

Hacking, Ian - Statistical and Inductive Probabilities (Review) (1964)

Uploaded by

themistoclesxii
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

BOOK REVIEWS 281

p. 2-9, bottom line, ' m \-n ' should be ' m \-m ' ; on p . 45, theorems 2.41 and 42 simply
reproduce 1.10 and 11, and should probably be replaced by their " imported " forms.
A. N. PRIOR

Statistical and Inductive Probabilities. By HUGTJES LEBLANC. (Englewood Cliffs, N . J . :


Prentice-Hall. 1962. P p . xii + 148. Price $6.65, or classroom edition $5.00).

According to the preface, this book tries to settle the dispute between followers
of von Mises and of Keynes. The former are said to study " statistical probabilities ",
the latter, " inductive " ones. F a r from being unrelated, inductive probabilities " qualify
as estimates of " the statistical kind. Leblanc's account of estimation is too vague for
this reconciliation to work, b u t the book is more modest than its preface. I t does little

Downloaded from http://pq.oxfordjournals.org/ at New York University on May 8, 2015


more than connect two definitions, and incidentally tell how to construe probabilities
as t r u t h values. There is little philosophical discussion.
Statistical probabilities are to be real numbers assigned to subsets of a given set in
accordance with Kolmogoroff 's axioms. For a finite set this means that the numbers
lie between 0 and 1, t h a t 1 is assigned to the whole set, and that the number assigned
to the union of two disjoint subsets is the sum of the numbers assigned to each. Sta-
tistical probabilities can be applied to sentences in a formal system. For ease of ex-
ample consider a language which names finitely many things. If there is an assignment
of statistical probabilities to all subsets of named things, the statistical probability
assigned to the open sentence ' w outlives Peter ' is the statistical probability of the
set of named things which do in fact outlive Peter. A closed sentence is assigned 1 or
0 according as it is true or false. Leblanc calls the statistical probability of any sentence
its truth-value.
As for Carnap, inductive probabilities are numbers between 1 and 0 assigned to pairs
of sentences in a formal language. If the first member of the pair is an hypothesis,
and the second some evidence, the inductive probability is 1 if the evidence entails
the hypothesis. Urging his main thesis, Leblanc says the inductive probability of h
on e qualifies as a fair or better estimate, made in the light of e, of the statistical pro-
bability, or t r u t h value, of h.
Leblanc is generous. An assignment of inductive probabilities is to be an estimator
estimating the truth-value of any h on any e. Assume both h and e are contingent.
Then, essentially, an estimate is " fair or better " so long as it gives 1 when e entails h,
0 when it contradicts h, and is such t h a t the estimates of h and r^h on e add up to 1.
If a coin were tossed a thousand times, falling heads just once, an estimate of the
probability of heads a t .99, and of tails at .01, would be " fair ". So stupid estimates
are " fair ". Distinguishing the fair from the better is called a " pressing issue ", but
the author says little about it.
Yet Leblanc is restrictive. Why must estimates of the t r u t h value of h and of '~ft.
add up to 1 ? There are at least some occasions when Wald's minimax rules seem
sensible. Then the best estimators are not additive. Carnap has argued, on obscure
grounds, t h a t estimators must be additive. Leblanc never mentions Wald, who now
dominates much statistical theory, nor says whether he likes Carnap's argument. Yet
additivity is the only substantial restriction on estimation, and estimation is supposed
to convey the main thesis of this book.
The gap is typical. There is no discussion of what makes an estimate good. Chapter
2 briefly mentions maximum likelihood estimates (which date from Daniel Bernouilli,
and have been systematically studied since Fisher's work in 1921) and says rather
scornfully t h a t these methods " are often said to have impressive credentials ". I n
fact, estimation has been closely analysed for years ; criteria for good estimates abound ;
their formal properties are catalogued in many textbooks; maximum likelihood methods
have been superseded by or contained in others. There is no whiff of this ferment in
Leblanc's book. Gauss and Laplace could measure the goodness of some estimates, but
not Leblanc. I t is now rather generally agreed that the excellence of an estimate de-
pends on the purpose to which it will be put. Leblanc doesn't mention the idea.
This book is a minor contribution to the study of formal languages. The exposition
is forthright. There is a dogged attempt to be lively by using fancy words—members
of a set " hail " from t h a t set ; one doesn't name a set but " dubs " it with that name.
Such phrases can be stunning if accompanied by the charm of a Quine, but here they
are tedious. Prentice-Hall are to be congratulated on their fine lay-out, which makes
a book full of logical notation very easy to read.
I A N HACKING

You might also like