Peircean Induction and the Error-Correcting Thesis
Mayo, 2005. §1, §2, §3, §4, §5, §6, §8, §9.
- Shelf
- Severity and error
- Concepts
- Severity, Convergent paths
Why it is hereThe paper through which the programme reads Peirce: induction is justified by its capacity to correct its own errors, and that capacity is measured as severity. Six of the programme’s claims come from Peirce by way of these pages.
The underlinings across the pages
pp. 298–317 · 20 marks
structure (red in Zotero) definition (green) key passage (yellow) book pages · pages not read are folded
1 Introduction
pp. 299–300
Peirce justified induction by its self-correcting character, and most of his readers, sympathetic or not, think the justification fails. Mayo argues it holds once it is read properly, and that it grounds frequentist statistics better than the statisticians' own rationale.
2 Probabilities are assigned to procedures not hypotheses
pp. 300–302
Like Neyman and Pearson, Peirce gives probabilities to methods, not to conclusions: how often a form of inference would lead us right. Priors over hypotheses are not statistical facts, and subjective ones only measure how a new idea fits our prejudices. Where Peirce goes beyond today’s error statistician is in asking what a test teaches in the case at hand, not only how it behaves in the long run.
C039 →Conceptual tool
3 So why is justifying Peirce’s SCT thought to be so problematic?
pp. 302–304
The critics read quantitative induction as the straight rule and qualitative induction as hypothetico-deduction, and ask a self-correcting method to converge on the truth and to supply better hypotheses mechanically. Both readings miss what Peirce meant: induction is reliable testing, and its worth is how well it uncovers mistakes.
4 Peircean induction as severe testing
pp. 304–306
For Peirce a test looks for consequences that would be surprising if the hypothesis were false, and a hypothesis that survives earns positive standing. That is severity: what is graded is not the support for H but the capacity of the test to catch the errors an inference to H declares absent.
- Testing, not agreement A test examines the consequences that would be unlikely were the hypothesis false, not how well the facts accord with it.C040 →Conceptual tool
- Peirce against Popper Popper keeps only refutation and will not call a well-corroborated hypothesis warranted. Peirce wants what a passed test positively tells us.
p. 304“Peirce, unlike Popper, is primarily interested not in falsifying claims but in the positive pieces of information provided by tests”
- The severity requirement H passes a severe test when the data accord with H and the procedure would very probably have signalled an error had H been false. The higher the probative capacity, the more severely H has passed.
- Two corrections to the critics Quantitative and qualitative induction differ only in whether severity can be measured or only judged. And the thesis has to justify accepting what passes, not only rejecting what fails.
5 The path from qualitative to quantitative induction
pp. 306–307
Peirce’s orders of induction are steps on one scale of severity, ranked by how much the confirmed prediction runs against what we would have expected without the hypothesis.
- First order: crude induction No evidence against H, so provisionally adopt H. It corrects itself only with a bang, and Peirce keeps it out of science: absence of counter-evidence is evidence only if the counter-evidence would have shown up.
- Second order: qualitative induction Fulfilled predictions, weaker when they only repeat what suggested the hypothesis, stronger when they are novel. Strength is judged, not measured.
- Third order: statistical induction When the surprise can be quantified, the test has an error probability and the induction becomes quantitative.
6 Quantitative and qualitative induction: significance test reasoning
pp. 308–311
A significance test is quantitative severity, and its self-correction is not climbing probabilities but checking whether an agreement recurs as often as chance would make it recur. Its assumptions are criticised by the same reasoning, informally; and data dredging wrecks severity, which is why randomisation and predesignation are rules of induction.
- Quantitative severity If the null were true, a result this significant would rarely occur; when it does, the alternative passes with severity one minus the p-value. A wrong rejection shows itself because the result fails to recur.C041 →Working method
- Semi-formal or qualitative severity Observational studies of hormone replacement therapy could not separate its effect from the advantages of the women who took it, so agreement with the hypothesis of a benefit was no severe test. The strongest severity arguments need no formal assessment.
- Predesignation Testing the one factor out of twenty that looked impressive turns a nominal 5% into roughly 64%: the selection belongs to the test.
8 Random sampling and the uniformity of nature
pp. 314–316
Severity needs a link between data and the process that produced them, but not the uniformity of nature. Randomness need only be approximate, departures from it weaken a conclusion without destroying it, and different methods of selection that err differently and still agree point to the truth.
- The link, not uniformity The justification lies in how the data were obtained: methods with good error probabilities, which check their own assumptions.
- Different selections If two selections made by different methods give nearly equal ratios, they are presumably near the truth.C042 →Conceptual tool
- No malign demon Induction only assumes that nothing systematically thwarts chance, and lesser biases are themselves detectable.C043 →Conceptual tool
9 Conclusion
pp. 316–317
The scepticism about the self-correcting thesis comes from a wrong picture of inductive testing and of what self-correction requires. Showing that induction corrects itself means showing that severe tests exist and that they let us learn from error.
- The guarantee Inductive methods are justified to the extent that they correct their own errors.
p. 316“inductive methods - understood as methods of severe testing - are justified to the extent that they are error-correcting methods”
Claims drawn from this card
Each claimA sentence from a source, kept verbatim, that the programme stands on, uses or has to answer to. shows its ENTIAThe notation this site uses on itself: what each piece is for, read through five layers. functionWhat a claim does for the programme: a conceptual tool, a working method, corrective evidence, a technical reference.; open + for its stanceHow the programme holds the claim: assumed, used, or contested., claim typeWhat kind of statement it is, so that the right test can be applied. Proposed until checked by hand. and relationsHow one claim bears on another, in the CiTO vocabulary: refines, extends, presupposes, confirms, contests..
+
used D1.6 interpretive-hermeneutic+
assumed D1.14 metaphysical-ontologicalParaphrase CC BY-NC 4.0, Claudio Cammarano. Quoted passages belong to their author and are kept short; the text itself is not distributed. Notes from the Zotero library, updated weekly.