Book contents
- Frontmatter
- Contents
- Acknowledgments
- Chapter 1 The paradox of predictivism
- Chapter 2 Epistemic pluralism
- Chapter 3 Predictivism and the Periodic Table of the Elements
- Chapter 4 Miracle arguments and the demise of strong predictivism
- Chapter 5 The predicting community
- Chapter 6 Back to epistemic pluralism
- Chapter 7 Postlude on old evidence
- Chapter 8 A paradox resolved
- Glossary
- Bibliography
- Index
Chapter 7 - Postlude on old evidence
Published online by Cambridge University Press: 22 September 2009
- Frontmatter
- Contents
- Acknowledgments
- Chapter 1 The paradox of predictivism
- Chapter 2 Epistemic pluralism
- Chapter 3 Predictivism and the Periodic Table of the Elements
- Chapter 4 Miracle arguments and the demise of strong predictivism
- Chapter 5 The predicting community
- Chapter 6 Back to epistemic pluralism
- Chapter 7 Postlude on old evidence
- Chapter 8 A paradox resolved
- Glossary
- Bibliography
- Index
Summary
INTRODUCTION
In 1927 Bertrand Russell delivered a characteristically subversive lecture entitled “Why I am not a Christian.” In 1980 Clark Glymour published an essay with the clever title “Why I am not a Bayesian.” Glymour's critique of Bayesianism resembled Russell's critique of Christianity in arguing that it was a view that was far more popular than it should be, given its merits. Among the various problems suffered by Bayesianism was one Glymour called the problem of old evidence.
To say that evidence e confirms hypothesis h on a Bayesian analysis is, intuitively, to say that p(h/e) > p(h). However, when e is antecedently justified (i.e. when e is ‘old evidence’) and p(e) = 1 it follows that p(h/e) = p(h) (assuming e is a logical consequence of h). Thus it appears that old evidence cannot confirm a hypothesis and – glory be – predictivism is vindicated! Not only predictivism, but the null support thesis – the radical Popperian view that old evidence is powerless to confirm a theory – seems to be straightforwardly established. But a moment's reflection suffices to realize that we have proved too much – for it is incontestable that old evidence can offer confirmation of a theory, as we saw, for example, in the case of Mendeleev's periodic law. But then it appears that, from a Bayesian point of view, the inequality p(h/e) > p(h) does not capture what it means to say that e confirms h for old evidence e.
- Type
- Chapter
- Information
- The Paradox of Predictivism , pp. 217 - 239Publisher: Cambridge University PressPrint publication year: 2008