I have mentioned Mayo and her book, STATISTICAL INFERENCE AND SEVERE TESTING before. Here I come back to it to provide you with some fruits of my readings therein.
Duhem's problem. Pierre Duhem said that experimental falsification doesn't get us very far in physics because the error that explains the surprising experimental result could invariably have come from any of a number of related postulates. The surprising result only tells us that ONE of them is wrong.
"The only thing the experiment teaches us is ... there is at least one error; but, where this error lies is just what it does not tell us." [a 1954 book.]
Quine said much the same, though he didn't think the idea restricted to physics.
Popper replied, in CONJECTURES AND REFUTATIONS (1962). "We can be reasonably successful in attributing our refutations to definite portions of the theoretical maze. (For we are reasonably successful in this -- a fact which remains inexplicable for one who adopts Duhem's and Quine's view of the matter.)"
Mayo takes Popper's side here, with an important qualification. Interestingly, she names some probability statisticians who were "ahead of Popper," and on that list is one of the classic American pragmatists, Charles Peirce. She gives nothing specific justifying the Peirce reference here, though there are -- says the index -- going to be some later mentions of Peirce.
Here is Mayo's qualification.
"Popper talked of wanting to subject theories to grave risk of falsification. I suggest that it's really our inquiries into, or tests of, the theories that we want to subject to grave risk. The onus is on interpreters of data to show they are countering the charge of a poorly run test. I admit this is a modification of Popper. One could reframe the entire demarcation problem as one of the characters of an inquiry or test."
In other words, she wants to go "meta" on the concept of falsification. Take an astrologer's hypothesis, "A Libra is more likely to become a successful investor than a Gemini." It is possible we could specify that statement with enough precision to make it falsifiable by proper statistical evidence. It is, though, not especially plausible to think that astrology becomes a science rather than a pseudo-science by allowing for that possibility.
A better inquiry: do astrologers routinely DO rigorous statistical inquiries into the truth of such statements? If so, then one astrologer might say to another, perhaps through the medium of a peer-reviewed journal, "your numbers on the Libra/Gemini investment disparity represent too small a data set. I have expanded the data set and obtained different results." Such exchanges, and the accompanying possibility that statistical rigor might break out, are a bit like the Loch Ness monster, and the accompanying possibility that some harmless Scot may end up in the belly of the beast.
On the other hand, it would be a matter of routine for astronomers to inquire into a statement about how one class of stars is more likely to go supernova than another. Routine, too, for this to become a dispute about the size of the data set. That is the new "demarcation" that Mayo proposes. It is not the theories but the tests of the theories that must be open to challenge and refutation.
This leaves us with the Duhem problem though.
Comments
Post a Comment