{"path":"research/burden-of-proof-and-priors/psychology-and-cases-report.md","content":"# Who must prove, in psychology and in real cases: 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. Nothing below was\nread in full text; the tags say how far each source was read, and \"[unverified]\" marks memory.*\n\n---\n\nRESEARCH REPORT: who must prove, in psychology and in real cases\n\nMethod and limits. I used WebSearch only. Search returned summaries and abstracts, not full texts, so no item below is [full text]. Tags used: [abstract], [secondary], and [unverified] for things I know from memory but could not confirm on the web. I give no effect sizes for the lab studies because I could not retrieve them. Where a number appears, it is in the cited source.\n\nPART 1 - THE PSYCHOLOGY OF WHICH SIDE MUST PROVE\n\na. Ditto & Lopez 1992, \"Motivated skepticism,\" JPSP 63(4): 568-584. Full PDF exists at https://fbaum.unc.edu/teaching/articles/jpsp-1992-Ditto.pdf, which I did not read. I only saw a search summary [secondary/abstract].\n- Study 1: participants judged which of two students was more intelligent. They needed less information to decide that a dislikable student was less intelligent than to decide he was more intelligent.\n- Studies 2 and 3: a fake medical test. People who got an unfavorable result took longer to decide the result was final. They were more likely to retest, cited more \"life irregularities\" that might have spoiled the test, and rated the test as less accurate.\n- Study 3 was added because Study 2 was confounded: illness is rarer than health, so the result could reflect surprise rather than preference. Study 3 led all participants to expect irregularities. The motivational account survived.\n- The authors' conclusion: the difference lies in how much processing preferred and unwanted conclusions get, so the bar is higher for the unwanted one.\n- Not found: effect sizes and sample sizes. A secondary account of the strip-waiting task describes people hoping for a positive result waiting longer. I could not confirm that design in this paper.\n- Note on wording: this is \"higher decision criterion\" for the unwanted conclusion, not a direct measure of acceptance rates.\n- Link for Deliberus: this is a clean empirical model of burden allocation. The burden goes to whoever holds the unwanted conclusion.\n\nb. Gilovich 1991 and Dawson, Gilovich & Regan 2002.\n- The \"Can I believe this?\" versus \"Must I believe this?\" framing is Gilovich's. I did not verify the wording on the web [unverified].\n- Dawson, Gilovich & Regan, \"Motivated reasoning and performance on the Wason selection task,\" PSPB 28: 1379-1387 [abstract]. In two studies, participants who got a rule implying their own early death (Study 1) or the truth of a threatening stereotype (Study 2) did far better than participants given non-threatening or agreeable rules. Reason: the right strategy on the task is to look for disconfirmation, and motivated rejection triggers that. I could not verify sample sizes or effect sizes.\n- Takeaway: unwanted claims get a stricter test, and that stricter test can improve accuracy. Skepticism is not always a bias. Its direction is what is biased.\n- Source: https://journals.sagepub.com/doi/10.1177/014616702236869 (the search surfaced this link, but a Sage ID under 0146167202... suggests the PSPB paper; confirm).\n\nc. Taber & Lodge 2006, AJPS 50(3): 755-769 [abstract/secondary].\n- Two experiments on affirmative action and gun control. People rated arguments matching their view as stronger, counter-argued opposing arguments, and chose confirming sources when free to choose. Attitude polarization was strongest among those with strong priors and high political sophistication.\n- Replication status: I found no direct replication of the prior-attitude effect. I found a critical review by Druckman (2012), which argues the study is internally sound but that motivated reasoning may be overstated outside the lab. The Guay & Johnston 2022 replication I found concerns a different study (asymmetric motivated numeracy) and found it robust [secondary]. Do not conflate the two. Your corpus already holds the Bullock and Tappin papers, which bear on this.\n\nd. Self-affirmation, with honest replication status.\n- I could not retrieve Cohen, Aronson & Steele 2000 or Sherman & Cohen 2006 on the web [unverified; the original findings are from memory].\n- Evidence on how well it holds up:\n  - Lyons et al. 2021, Journal of Experimental Political Science 9(2): 225-240 [abstract]: studies from several research teams, using conceptual extensions rather than close replications. They tested political attitudes, factual beliefs, conspiracy beliefs, affective polarization, and evaluation of news sources. The teams consistently found that self-affirmation had little effect. https://ideas.repec.org/a/cup/jexpos/v9y2022i2p225-240_6.html\n  - Hanselman et al. (Journal of Educational Psychology 109(3)) [abstract/secondary]: a well-powered replication in education, in the setting of an earlier large positive field experiment, found no effect, with estimates precise enough to rule out benefits above d = 0.10. Moderators theorized to explain the difference did not account for it. The WUSTL profile page names this as \"New evidence on self-affirmation effects and theorized sources of heterogeneity from large-scale replications.\"\n  - The same sources cite a counterpoint, Borman et al. 2018 [secondary].\n  - A 2016 Minnesota dissertation meta-analysis of defensive processing in health messages (55 studies, N = 10,611) found small positive effects [secondary, a dissertation, not peer reviewed].\n  - I found no multi-lab registered report specific to defensive processing of threatening evidence.\n- Verdict: for political and factual belief, self-affirmation should be treated as unreliable. For health messages, small effects are reported. It is not a safe design lever for Deliberus.\n\ne. Abelson 1986 and Kaplan, Gimbel & Harris 2016.\n- Abelson, \"Beliefs are like possessions,\" J. Theory of Social Behaviour 16(3): 223-250. Core claim: for many people, beliefs function like possessions, and reasoned argument often fails to change them [secondary]. A full-text PDF is at gwern.net/doc/sociology/1986-abelson.pdf, not read.\n- Kaplan, Gimbel & Harris, Scientific Reports 2016 (doi 10.1038/srep39589) [abstract/secondary]: 40 liberal participants, so generalization to others is limited. Challenges to political beliefs raised activity in the default mode network. Participants who changed their minds more showed less amygdala and insula activity. Political beliefs moved less than non-political beliefs. The \"emotion drives resistance\" reading is the authors' interpretation (reverse inference from brain regions). The study is a single small fMRI sample, not replicated as far as I found.\n\nf. Status quo bias, Samuelson & Zeckhauser 1988, J. Risk and Uncertainty 1(1): 7-59 [abstract/secondary].\n- Hypothetical experiments and field data (faculty choosing health plans and retirement programs) showed disproportionate sticking with the status quo. Explanations come from economics, psychology and decision theory.\n- Replication: Xiao et al., Metapsychology, retested four scenarios. Three supported the bias, and the wagon-colour scenario did not [secondary].\n- Link for Deliberus: a default position is an implicit burden rule. Whoever proposes a change must overcome inertia.\n\ng. Testimony, conformity, and shared sources.\n- Bikhchandani, Hirshleifer & Welch 1992, J. Political Economy 100(5): 992-1026 [abstract]: in an informational cascade, it is optimal to follow those ahead regardless of your own information. One later note says the result depends on a tie-breaking convention.\n- Bovens & Hartmann 2003, Bayesian Epistemology (OUP) [secondary]: agreement among sources adds support only to the extent that sources are independent. If all copy one report, the group adds no more than that one report. This is the formal treatment you asked for. A critique reports that more independence does not always mean stronger confirmation, and that the variety-of-evidence thesis holds up when unreliable sources are modelled as systematically biased.\n- Not found on the web: Asch figures, Ross, Greene & House 1977 on false consensus, and Easwaran et al. \"Updating on the credences of others.\" I did not verify any numbers for these. Do not quote an Asch percentage from this report.\n- Design implication: count a claim's supporters by distinct origin, not by number.\n\nh. Bias attributions as cheap and symmetric.\n- Pronin, Gilovich & Ross 2004, Psych Review 111: 781-799 [abstract/secondary]: people see others as more biased than themselves, tied to naive realism and to overweighting one's own introspection.\n- Kennedy & Pronin 2008, PSPB 34: 833-848 [secondary]: in disagreements, people think the other side is biased, and this feeds conflict escalation.\n- Gap: I found no study showing that bias accusations predict actual reasoning quality or accuracy. The evidence I found shows that accusations track disagreement, not the opponent's reasoning. Treat \"does a bias accusation predict anything?\" as unanswered by this search. It needs a targeted search if you want it.\n\nPART 2 - REAL CASES\n\n1. Thalidomide and Frances Kelsey (FDA, 1960-62).\n- Burden: on the manufacturer, Richardson-Merrell (Kevadon). Kelsey was the FDA reviewer.\n- Evidence: the company's data were largely physician testimonials, with too few clinical studies and little pregnancy-safety data. She repeatedly asked for more. The company withdrew in April 1962 [secondary: ASU Embryo Project, Smithsonian, UChicago, PolitiFact].\n- Pre-approval distribution had already put over 2.5 million tablets with 1,200+ physicians and about 20,000 patients in trials [secondary, same sources].\n- Outcome: the US largely avoided the birth-defect epidemic seen elsewhere. The 1962 Kefauver-Harris Amendments required proof of effectiveness, not only safety, before marketing.\n- Verdict: the burden on the maker helped. Caveat: sources differ on some details, and I did not check the statute.\n- Source: https://embryo.asu.edu/pages/us-regulatory-response-thalidomide-1950-2000\n\n2. Semmelweis and handwashing (Vienna, 1847-).\n- Burden: de facto on Semmelweis, an outsider challenging senior clinicians such as his superior Johann Klein.\n- Evidence: doctors' ward mortality about three times the midwives'. In April 1847, before the new procedure, mortality was 18.3%. After chlorinated-lime handwashing began in mid-May: June 2.2%, July 1.2%, August 1.9% [secondary: Wikipedia, MJA]. Numbers are approximate.\n- Outcome: his ideas were rejected by the medical community. He was committed to an asylum in 1865 and died there; the search did not state the death explicitly, so that is from memory.\n- Verdict: the burden hurt. The data were strong, but there was no accepted mechanism (germ theory came later), so the demand for a mechanism blocked the intervention. I could not confirm Skoda's or Pasteur's roles.\n- Source: https://www.mja.com.au/journal/2018/209/11/ignaz-philipp-semmelweis-1818-1865-herald-hygienic-medicine\n\n3. Wegener and continental drift (1912 to the 1960s).\n- Burden: on Wegener. The critics' objection was mechanism: his idea that continents \"plow\" through ocean floor was held physically impossible [secondary: USGS].\n- Evidence: Hess 1962 proposed seafloor spreading, which removes the plowing objection. Vine and Matthews (1963) argued the new crust would record magnetic reversals \"like a tape recorder.\" The Morley manuscripts were rejected by Nature and JGR [secondary].\n- Outcome: acceptance came through these later arguments. I did not confirm a specific year for the shift.\n- Verdict: the burden was partly legitimate (no mechanism) and partly obstructive (even supporting evidence found it hard to publish). Mixed.\n- Sources: https://pubs.usgs.gov/gip/dynamic/HHH.html\n\n4. Marshall & Warren, H. pylori (1982-2005).\n- Burden: on the discoverers, against the settled view of ulcers.\n- Evidence: Warren saw spiral bacteria in biopsies. Culture succeeded by accident in 1982 when plates were left over Easter. Marshall drank a culture and developed gastritis. Sources disagree on the date (1984 or 1985) [secondary].\n- Outcome: initial skepticism. The NIH endorsed antibiotics for ulcers in 1994 [secondary]. Nobel Prize 2005 [Nobel committee speech, https://www.nobelprize.org/nobel_prizes/medicine/laureates/2005/presentation-speech.html].\n- Verdict: roughly a decade to move the burden. The delay was real but ended with evidence, and the case shows that even correct claims can need heroic proof.\n\n5. Cold fusion (1989).\n- Burden: on Fleischmann and Pons, treated as an extraordinary claim.\n- Evidence: failed replications at the May 1989 APS meeting; a November 1989 DOE panel found no convincing evidence but left normal research channels open [secondary: Scientific American, Skeptic]. The non-replications were themselves criticized (MIT work faulted for data reduction), and later groups claim excess heat. A 2004 IEEE Spectrum piece notes excess heat alone does not prove fusion.\n- Verdict: rejection was probably right on the fusion claim, but the process was fast and partly rhetorical (Koonin's \"incompetence and delusion\" remark is reported secondhand). I found nothing that frames it explicitly as burden of proof.\n\n6. Bem's precognition (2011) and replications.\n- Burden: on Bem, with high prior skepticism because the claim is physically implausible.\n- Evidence: nine experiments in JPSP. Ritchie, Wiseman & French 2012 (PLoS ONE): three pre-registered replications of one experiment, combined n = 150, all failed [abstract]. Galak, LeBoeuf, Nelson & Simmons 2012 (JPSP 103: 933-948): seven experiments, N = 3,289, replicating Bem's experiments 8 and 9; meta-analytic effect d = 0.04 [abstract].\n- Verdict: a high prior plus replication worked. Note the journals accepted the paper but critics then applied the higher bar.\n- Source: https://pmc.ncbi.nlm.nih.gov/articles/PMC3303812 and https://papers.ssrn.com/abstract=2001721\n\n7. Tobacco \"doubt is our product\" (1969).\n- Burden: the industry's strategy was to make the public health side prove causation to a standard the industry would never accept.\n- Evidence: a Brown & Williamson \"Smoking and Health Proposal,\" doc no. 680561778-1786, 1969, widely quoted as saying doubt is the industry's product. Attribution to an author is a guess (SourceWatch). Wikipedia attributes it to R. J. Reynolds, which conflicts [secondary]. I did not read the primary memo; it is in the UCSF Legacy library.\n- Verdict: burden-shifting was used as a weapon. I have not verified the exact wording here, so do not quote it from this report.\n- Sources: https://sourcewatch.org/index.php/Smoking_and_Health_Proposal\n\n8. Leaded petrol and the Kehoe paradigm.\n- Burden: Kehoe (1925) required evidence of harm before restriction (\"show me the data\"), with a cascading uncertainty rule and skewed cost-benefit weighting [secondary: Nriagu 1998, Environmental Research]. A commonly cited formulation is \"in the absence of clear evidence of risk there is no risk of significance\"; secondary sources vary in wording.\n- Evidence: Patterson showed average Americans carried about a hundred times natural lead levels [secondary].\n- Outcome: removal began with the Clean Air Act of 1970. Sources give different end dates (1986 versus 1996), reflecting different milestones.\n- Verdict: the burden on regulators hurt, for decades.\n- Source: https://litfl.com/precautionary-principle-and-the-kehoe-principle/\n\n9. O. J. Simpson (criminal 1995, civil 1997).\n- Burden: prosecution beyond reasonable doubt (acquitted 3 October 1995). Plaintiffs on preponderance in the civil case, which found him liable with $33.5 million in damages [secondary: CNN, Britannica].\n- Evidence: the defense in the criminal case focused on contamination and planting allegations. The civil case included material the criminal jury did not see, such as photographs of him in shoes matching bloody footprints [secondary, one source, unconfirmed against better ones]. So \"same evidence\" is not accurate here.\n- Verdict: the standard affects the outcome, but this case is confounded. Use cautiously as an illustration of standards, not a pure test.\n\nRANKING FOR USE\nStrongest cases for burden-allocation arguments: thalidomide (burden helped), Kehoe and tobacco (burden hurt), Semmelweis (a mechanism demand hurt). Bem is the best case for a justified high prior. Simpson is the weakest.\n\nKEY GAPS\n- No effect sizes or full texts for Part 1.\n- Self-affirmation original studies and the Cohen et al. 2000 replication status not verified.\n- No evidence found on whether bias accusations predict anything.\n- Asch and false consensus figures not verified.\n"}