[LukasWallrich]

Description

Cross-checking the Replications and Reversals page against FLoRA and the cited papers turned up entries that misreport their own sources. For example, the page lists the bottomless soup bowl as "not replicated", although a preregistered direct replication (Lopez et al., 2024, n = 464) reproduced the effect at about half the original size. In another entry, the hurricane-names percentages are male/female ratios mislabelled as decreases. This PR corrects only the errors confirmed against the primary sources. The evidence for each change, with short quotes and page numbers, is in the corrections ledger.

Status changes

Two rules: reversed needs a statistically supported effect in the opposite direction, and a null original has no direction to reverse.

• Bottomless soup bowl: not replicated → replicated (about half the original effect size). Adds Lopez et al. 2024 and keeps the critical blog posts.
• Three entries whose original finding was null move from reversed to not replicated (…): age of acquisition (AoA) in recognition, the AoA verification-task entry, and numerical ratio effects in elephants (no ratio effect in Asian elephants; a later ratio effect in African elephants). The sibling AoA object-decision entry already uses this wording.
• Three entries whose follow-ups found only non-significant opposite-sign estimates move from reversed to not replicated (…): hurricane names (Smith 2016, p = .31 and .50), elderly priming (Doyen et al. 2012: no effect with automated timing unless experimenters expected slower walking) and scarcity resource allocation (O'Donnell et al. 2021: F(1, 530) = 3.59, p = .059).
• Depressive realism: reversed → mixed (much smaller effect in the original direction). The meta-analysis reports d = −0.07 in the same direction.
• Object distance: reversed → NA. The only cited follow-up used a different task, so it cannot reverse the original effect.

Labels checked because FLoRA disagreed

Where FLoRA's replication codes disagreed with a page label, both were checked against the papers. The FLoRA code alone never decided a label.

• Abstract rule learning, induced-compliance dissonance, choosing vs rejecting, facial feedback, moral priming (cleanliness) and men's preference for competition become mixed. Each now cites the replication the page had omitted: Geambașu et al. 2022, Vaidis et al. 2024, Many Labs 2 and Shafir & Cheek 2024, Many Smiles, Huang 2014 and Carlsson et al. 2020. With the original materials, Many Labs 2 found a significant opposite effect for choosing vs rejecting.
• Scarcity brand attitudes ("replicated (NS)…") and neonate imitation (NA) become not replicated (…). Neonate imitation now also cites Koepke et al. 1983.
• Last place aversion: replicated → mixed. Two direct replications found no last-place aversion: Camerer et al. 2016 (the redistribution game) and Martinangeli & Windsteiger 2021. A modified design and a queueing study found support. The description now covers both of the original's experiments, not just the gambles. The original's DOI pointed to an unrelated chemistry cover image and is fixed.
• Prosocial spending: replicated → replicated (smaller effects; one of three experiments was null).
• Less-is-better: fixes "d = 76" and a confidence interval.
• FLoRA's own errors are filed as FLoRA rows to correct, found while cross-checking the Reversals page fred-data#175.

Description, sample and study-type fixes

• Elephants: the original tested five elephants in Experiment 1 and four in Experiment 2, not three; effect lines say which condition the correlations come from.
• Hurricane names: describes Smith's data correctly (expanded Atlantic set and a new Pacific set, not "same data") and replaces the impossible "264%/103% decrease" with the reported means and p values.
• Object colour: describes Connell's original mismatch advantage.
• Facial feedback: separates the Coles et al. 2019 meta-analysis (286 effect sizes, 138 studies) from the Many Smiles study (n = 3,878).
• Loss aversion: removes a misplaced unconscious-thought meta-analysis.
• Slow blinking: corrects the sample to n = 18.
• Colour approach/avoidance: marks the 2013 poster as the same sample as Steele 2014.
• Peptic ulcers: identifies Levenstein et al. 1999 as a review; the page had attributed an n = 4,595 sample to it.
• 5-HTT × stress: Chabris et al. 2012 (intelligence) and Farrell et al. 2015 (schizophrenia) test other genes, so they move from the replication line to a labelled context note. Serretti et al. 2007 (HTR2A, unlinked) is removed.

Links

• 22 citations (25 links) pointed to unrelated works. 19 now point to the intended work, each checked on Crossref against the cited authors, year and title. Three are unlinked because the intended identifier could not be verified.

The remaining six reversed labels (Romeo and Juliet, monolingual creativity, object colour, handedness and dyslexia, mate guarding, personality and political attitudes) are kept; the ledger lists how each was checked. No unrelated wording or citation formatting changed.

Related to #612.

Type of Change

• Content/documentation update
• New feature
• Bug fix
• Breaking change

Testing

• Tested locally: hugo build, scripts/check_seo_html.py and scripts/check_sitemap_lastmod.py pass; the edited entries render correctly
• Previewed on staging.forrt.org
• Not tested yet

Checklist

• Self-reviewed my changes
• Verified links and formatting are correct
• No new warnings or errors

Notes

git merge-tree shows no conflict with rebrand-doors.

Created with Claude Opus 5.5 in Claude Code (T3 Code); GPT-6.1 Sol independently reviewed the age-of-acquisition verification-task and elephants corrections, and the FLoRA disagreements (four reviews and a second review of the labels where we disagreed).

🤖 Generated with Claude Code

[github-actions]

👍 All image files/references (if any) are in webp format, in line with our policy.

[cubic-dev-ai]

No issues found across 1 file

_{View guided diff | Turn on auto-fix | Re-trigger cubic}

[github-actions]

📝 Spell Check Results

Found 1 potential spelling issue(s) when checking 1 changed file(s):

📄 content/reversals/reversals.md

Line	Issue
2243	acn ==> can, acne

---

ℹ️ How to address these issues:

1. Fix the typo: If it's a genuine typo, please correct it.
2. Add to whitelist: If it's a valid word (e.g., a name, technical term), add it to .codespell-ignore.txt
3. False positive: If this is a false positive, please report it in the PR comments.

_{🤖 This check was performed by codespell}

[LukasWallrich]

✅ Staging Deployment Status

This PR has been successfully deployed to staging as part of an aggregated deployment.

Deployed at: 2026-11-13 12:30:41 UTC
Staging URL: https://staging.forrt.org

The staging site shows the combined state of all compatible open PRs.

[cubic-dev-ai]

2 issues found across 1 file (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="content/reversals/reversals.md">

<violation number="1">
P2: The 'not significant' claim for O’Donnell et al.’s r = -.082 does not follow from the listed n = 573: that gives p ≈ .049, which is nominally significant. Confirm the p-value or analysis sample against the primary paper before keeping 'not significant' here, since the status label depends on it.</violation>

<violation number="2">
P3: Scope the automated-timing null to Doyen's first experiment. Their later expectancy experiment found slower walking in the expected-slow condition, including in its objective-timing data.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

View guided diff | Turn on auto-fix | Re-trigger cubic</sub>

[cubic-dev-ai]

1 issue found across 1 file (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="content/reversals/reversals.md">

<violation number="1">
P2: The Huang 2014 effect sizes are attributed to 'Study 2', but the critique line also lists a direct replication, Study 2a (n = 436), with no reported result. Report Study 2a's effect size, or state which sample the d values come from, before relying on this for the 'mixed' status.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

View guided diff | Turn on auto-fix | Re-trigger cubic</sub>

[LukasWallrich]

Note

🤖 Claude Opus 5.5 responding on behalf of LukasWallrich

Addressed in 5e57014. Two of the three points led to changes; the Huang one needed none.

• Scarcity: resource allocation. "Not significant" stands. O'Donnell et al.'s conversion code (osf.io/gra32) gives the replication test as F(1, 530) = 3.587, N = 573. That is p = .059: the page's n = 573 is the sample size, not the degrees of freedom of the test. The entry now shows the statistic.
• Elderly priming. Agreed. The status now reads "not replicated (no effect with automated timing unless experimenters expected slower walking)". This covers Doyen et al.'s Experiment 2, where objective timing showed slowing only in the slow-expectation condition.
• Huang 2014. No change. The effect line already says which sample it reports: "Huang, Study 2: low response effort, d = -0.29, p = .04; high response effort, d = 0.19, p = .16". Study 2a is listed as the direct replication supporting the same moderation.