'PHILOSOPHICAL LECTURES' SYMPOSIUM

Bayes's Theorem

Saturday 10 March 2001
The British Academy, 10 Carlton House Terrace, London, SW1Y 5AH

Bayes's Theorem is a powerful but controversial tool for assessing how probable specified evidence makes some hypothesis.  It claims that the probability of a hypothesis h on evidence e and background knowledge k (the posterior probability of h) is equal to the probability of e given h and k (the predictive power of h), multiplied by the probability of h given only k (the prior probability of h), divided by the probability of e given only k (the prior probability of e). The prior probability of e is its probability if h is true is multiplied by the prior probability that h is true, plus its probability if h is false is multiplied by the prior probability that h is false.

Representing by P(h/e & k) the posterior probability of h, by P(e/h & k) the predictive power of h, by P(h/k) the prior probability of h, and by  (e/k) the prior probability of e, Bayes's Theorem then reads

P (h/e & k) = 

 P(e/h & k) P(h/k)
______________________________________

 

P(e/k)

Thus suppose that we have background knowledge k that a certain coin has with equal probability either a bias of 2/3 in favour of landing heads, or a bias of 2/3 in favour of landing tails. Let h be the hypothesis that it has a bias of 2/3 in favour of landing heads.  Suppose we toss it four times and all these tosses are heads (e).  Bayes's Theorem then tells us that this evidence gives a probability of 16/17 to the hypothesis h.

We may not be able to give exact numerical values to the terms on the right-hand side of Bayes's Theorem, but so long as we can give rough values, we can give a rough value to the left-hand side. So perhaps we can use it to assess (roughly) the probabilities, not merely of simple statistical hypotheses, but of scientific theories, historical claims and world-views. Yet to apply it at all, we need to be able to ascribe prior probabilities to hypotheses.  But can we do that in an objective way, even when we are dealing with simple statistical hypotheses and have some substantial relevant background knowledge from observation?  And what about when all our observational evidence is included in e  - are there a priori criteria for ascribing prior probabilities? Or must the prior probability of a hypothesis measure merely a given person`s initial degree of confidence in that hypothesis? This symposium will investigate whether Bayes's Theorem has any philosophical justification at all, whether it has application in statistical science, in the law-courts, and in assessing the probability of world-views such as theism. There will be opportunity for discussion following each paper.

Attendance at the conference is free, and all those interested are welcome to attend, but it is essential to register in advance. In order to register, please complete the attached form and return it to Angela Pusey at the British Academy, 10 Carlton House Terrace, London SW1Y 5AH (telephone: 020 7969 5200; email: a.pusey@britac.ac.uk). If you would like to have buffet lunch at the Academy at a cost of £12 per person please indicate this on the form, and enclose a cheque made payable to the British Academy.


PROGRAMME 

Sat 10 March

 

 

 

10.00

Coffee and registration

Session I

10.30 

Introduction
Professor Richard Swinburne, FBA

 

11.00

Bayes's Theorem: The Philosophical Issues
Professor Elliott Sober, Professor of Philosophy, University of Wisconsin, Madison

 

12.15

Lunch

Session II:
Dr Jeremy Butterfield, FBA

13.30

Bayes's Theorem and Statistical Science
Professor Colin Howson, Professor of Philosophy, London School of Economics

Session III: 
Professor Hugh Mellor, FBA

14.45 

Bayes's Theorem  and Weighing Evidence by Juries
Professor Philip Dawid, Professor of Statistics, University College London

 

16.00

Tea

Session IV:
Professor Michael Redhead, FBA

16.30

Bayes's Theorem, Miracles and Theism
Professor John Earman, Professor of Philosophy, University of Pittsburgh


More about the Philosophical Lectures