{"path":"research/burden-of-proof-and-priors/theory-report.md","content":"# Burden of proof in theory: a research helper's report, verbatim\n\n*2026-10-23. Written by a research helper (Sonnet, web search only) for\n[../burden-of-proof-and-priors.md](../burden-of-proof-and-priors.md). Kept as returned. Its own warning\nstands: nothing below was read in full text; every tag says how far each source was read. Where it\nwrites \"from memory\", the claim is unverified.*\n\n---\n\nOverall: only abstracts and secondary pages were readable. Springer, Kaplan, Ullmann-Margalit, Walton 2014 and Rescher were all out of reach. Nothing below is [full text]. Where I fall back on memory, I say so.\n\n**A. Hahn & Oaksford 2007b, Argumentation 21: 39–61 (doi 10.1007/s10503-007-9022-6)**\n- The abstract says the burden of proof matters in law because law aims at practical decisions. In critical discussion the goal is \"frequently simply to increase or decrease degree of belief\", and there it is \"not necessarily important\" whether belief crosses a threshold such as reasonable doubt. The authors then ask what that means for the burden's role in argumentation and in theories of fallacy. [abstract] ([Birkbeck record](https://eprints.bbk.ac.uk/id/eprint/16126), [Springer](https://link.springer.com/article/10.1007/s10503-007-9022-6))\n- A secondary source (Goodwin-style ISSA 2010 paper) quotes them calling the carry-over from law a \"hasty transference\" to everyday argument, and says they cite Whately as the first to make it. [secondary] ([ISSA 2010](https://rozenbergquarterly.com/issa-proceedings-2010-burdens-of-persuasion-and-proof-in-everyday-argumentation/?print=pdf))\n- The companion Psychological Review paper (the one you already have) is quoted as rejecting burden-shifting as the explanation of why an argument from ignorance is weak. An argument fails its burden only if it is already weak, so the burden explains nothing. Strength depends on content, which their Bayesian account captures. [secondary] ([Hahn, Harris & Oaksford 2013](https://doi.org/10.1080/19462166.2012.689327))\n- On your question: the abstract does not say burden is meaningless without a decision. It does say the threshold, which is the part of burden that bites, matters mainly where there is a practical decision. Without one, the job is to move degrees of belief. The abstract does not use \"utility\". I could not verify any utility discussion, so I would not attribute one to them.\n\n**B. Walton's taxonomy**\n- The persuasion/production/tactical trio is not Walton's own. Burden of persuasion and burden of production come from evidence law (*McCormick on Evidence*). The tactical burden comes from Williams (2003), who calls it a \"merely tactical evaluation of who is winning\" at a given moment. Walton's own distinction (1988) is between global and local burdens. [secondary] ([ISSA 2010](https://rozenbergquarterly.com/issa-proceedings-2010-burdens-of-persuasion-and-proof-in-everyday-argumentation/?print=pdf))\n- Prakken & Sartor use this three-way split. The burden of persuasion says who loses if the evidence is balanced. The burden of production is the duty to put forward evidence on an issue. A presumption, once invoked, can shift the tactical burden to the other party. [secondary] ([Prakken & Sartor](https://webspace.science.uu.nl/~prakk101/pubs/jurix08pres.pdf), [CUP index](https://assets.cambridge.org/97811070/46627/index/9781107046627_index.pdf))\n- Walton 2014 (CUP) draws on how courts allocate presumptions and burdens. Its index lists an argumentation scheme \"argument from lack of knowledge\" at p. 193. I could not read how it treats the fallacy of shifting the burden or the argument from ignorance. [secondary]\n\n**C. Presumption theory**\n- Ullmann-Margalit 1983, J Phil 80(3): 143–63. Secondary sources confirm the formula Pres(P, Q): when fact P holds, act as if Q holds unless there is sufficient reason to think otherwise. She starts from legal presumptions and generalises to practical deliberation. [secondary] ([Doxa 21 PDF](https://rua.ua.es/dspace/bitstream/10045/10383/1/doxa21_05.pdf), [Wikipedia](https://en.wikipedia.org/wiki/Edna_Ullmann-Margalit)) My search and fetch did not surface the costs-of-error argument. From memory, I believe she justifies presumptions by probability and by the relative costs of the two errors, but I have not verified it.\n- Whately (1828): a presumption lies with any existing institution, and the party who disputes it carries the burden. Kauffeld (2003) criticised him for giving no criteria for identifying presumptions and for falling back on common sense. Others call his view special pleading, since he was an archbishop. [secondary] ([ISSA 2002](https://rozenbergquarterly.com/issa-proceedings-2002-the-ordinary-practice-of-presuming-and-presumption-with-special-attention-to-veracity-and-the-burden-of-proof/?print=pdf), [Godden & Walton](https://philarchive.org/rec/GODATO-3))\n- Godden & Walton describe presumption as a modal status attached to a claim that shifts a locally set burden. Rescher and Kauffeld/Freeman take the social, practice-based view. I could not verify the content of Rescher 2006 or the details of Kauffeld's pragmatic account. [secondary]\n\n**D. Standards of proof as thresholds**\n- Kaplan 1968, 20 Stan. L. Rev. 1065. Secondary sources say he treats conviction as rational when (1−P)·Di < P·Dg. P is the probability of guilt, Di the disutility of convicting an innocent person, and Dg the disutility of acquitting a guilty one. Solving gives threshold p* = Di/(Di+Dg). That last step is my own algebra, not quoted from him, and the formula matches your description. He did not mean it as a literal procedure. [secondary] ([Wikipedia on Kaplan](https://en.wikipedia.org/wiki/John_Kaplan_(law_professor)), [Quaestio Facti](https://revistes.udg.edu/quaestio-facti/article/view/23280))\n- Blackstone: \"It is better that ten guilty persons escape than that one innocent suffer.\" Variants are Hale 5:1, Fortescue 20:1, Franklin 100:1 and Maimonides 1,000:1. A 10:1 ratio gives p* ≈ 0.91 by the same algebra. Wikipedia reports Weinstein's roughly 90% for reasonable doubt. [secondary] ([Wikipedia](https://en.wikipedia.org/wiki/Blackstone%27s_ratio))\n- Critiques: Cullerne Bown argues the ratio is methodologically flawed. A 1996 Law & Social Inquiry article argues that, depending on how well a jury discriminates guilty from innocent, a probability threshold can produce almost any ratio of actual errors. [secondary] ([LSI 1996](https://www.cambridge.org/core/journals/law-and-social-inquiry/article/abs/difference-between-blackstonelike-error-ratios-and-probabilistic-standards-of-proof/10F0DD2777CDC5EC6B822C9D72EA195E))\n- Civil versus criminal: a Swiss empirical study is reported to find that judges' own error-cost estimates yield a threshold matching preponderance. I did not read the study. [secondary] ([MPG PDF](https://homepage.coll.mpg.de/pdf_dat/2013_12online.pdf))\n\n**E. Extraordinary claims**\n- Antecedents: Hume 1748 (\"Of Miracles\"), Laplace 1810/1814 (\"the more extraordinary a fact, the more it needs to be supported by strong proofs\"), Jefferson 1808, and Flournoy 1899. Truzzi used \"extraordinary claims require extraordinary proof\" in 1975 and 1978. The \"evidence\" wording became canonical around 1980. The Flournoy attribution rests on one source. [secondary] ([Quote Investigator](https://quoteinvestigator.com/2021/12/05/extraordinary/), [Wikipedia](https://en.wikipedia.org/wiki/Extraordinary_claims_require_extraordinary_evidence))\n- Bayesian reading: a low prior needs a high likelihood ratio to reach a high posterior. That is the usual reading, and it is also my own gloss.\n- Critique: \"extraordinary\" cannot be defined objectively and depends on one's worldview. Critics say it is invoked to dismiss anomalies and anything outside the mainstream. [secondary] (same sources)\n- Russell's teapot: \"the philosophic burden of proof lies upon a person making empirically unfalsifiable claims.\" [secondary] ([Wikipedia](https://en.wikipedia.org/wiki/Russell%27s_teapot)) I could not source Hitchens's razor. From memory it reads \"what can be asserted without evidence can also be dismissed without evidence\", from *God Is Not Great* (2007), but that is unverified.\n\n**F. Precaution and regulation**\n- Wingspread Statement (1998): \"the proponent of an activity, rather than the public, should bear the burden of proof\", and precautionary measures apply even if cause and effect is not fully established. [part] ([IATP](https://iatp.org/sites/default/files/Wingspread_Statement_on_the_Precautionary_Prin.htm)) I did not read the page itself, so the quote comes from a secondary summary.\n- REACH is described by the EEB and a regulator slide deck as \"no data, no market\", which puts the burden of proving safety on firms. The EEB says it is not working that way. ECHA issues registration numbers to incomplete dossiers, so the burden falls back on authorities. [secondary] ([EEB](https://eeb.org/en/?p=13362)) These are advocacy sources.\n- Sunstein: the precautionary principle is incoherent or paralysing because it condemns the very steps it requires, since regulation carries its own risks. I read this in the Daedalus preview of the book argument, not in *Laws of Fear* itself. [part] ([Sunstein paper](https://www.stat.berkeley.edu/~aldous/157/Papers/sunstein_prec.pdf)) I did not find anything on the proactionary principle.\n\n**Implications for an argument graph that separates (i) starting value, (ii) obligation to argue, (iii) threshold for acting**\n1. Hahn & Oaksford support keeping (iii) off the claim. Thresholds belong to decisions, and a graph that only moves belief has no use for one. This agrees with your rule 4.\n2. Kaplan's p* = Di/(Di+Dg) is a property of a decision, not of a claim. Store costs per decision context. The same claim at 0.8 can be enough for a civil decision and not enough for a criminal one.\n3. Burden of persuasion (who loses on a tie) is a decision-time tiebreak. Burden of production (who must speak next) is a dialogue or UI matter. Neither should write to the claim's value. Treat the tactical burden as derived state, not stored.\n4. Presumption in Ullmann-Margalit's sense is a rule that answers \"act as if Q\" and is justified partly by error costs. Keep it out of the prior. A presumption that moves a number is the Whately problem again, where common sense is dressed as a probability.\n5. The starting value (i) is where \"extraordinary claims\" lives. Its Bayesian reading is that the prior carries the work. The critique that \"extraordinary\" depends on worldview supports your reference-class proposal, because the class must be stated openly and be open to challenge.\n6. Teapot and Hitchens are rules about who must speak, i.e. (ii). They fit \"a reason that fails drops out\" and no push-down without an objection or a thorough search. Hahn & Oaksford's point that a burden explains nothing unless the argument is already weak is a caution against treating burden as a score.\n7. Precaution (Wingspread, REACH) reverses (ii) and (iii) together, by who must produce evidence and by a lower acting threshold. The graph should be able to record that as a decision-context setting without changing any claim's value. Sunstein's objection is the reason to expose the cost terms rather than hide them.\n"}