KurisutinaJournal of Cognition 4(1): 29, doi 10.5334/joc.161; open access (CC BY 4.0), PMC8194981. A Registered Report. Read by researcher R5a, 5 October 2026. Provenance: papers/change/henderson2021_truth_trajectory.provenance.json.
What was read
- Read in full: the Europe PMC XML, converted with
tools/pmc2txt.py(self-check passed). That covers:- abstract and introduction, including Table 1, which reproduces Dechêne et al.'s delay moderator;
- method, sampling plan and analysis plan;
- results with Tables 2–7, and the exploratory analyses;
- discussion, constraints on generality, footnotes, ethics, funding, contributions;
- the 57 references and the data statement.
- Not read: Figures 1–7 as images (the key values are in the text and tables); the OSF materials, online supplement and data.
What they did
- Sample. UK Prolific workers, English first language, aged 18–65.
- 631 recruited; 567 analysed at phase 1, 507 completed all four phases.
- Data collected December 2020 to January 2021, during COVID lockdown.
- Materials. 128 trivia statements, pretested so that their truth was uncertain (mean ratings 3.50–4.53 on a 1–7 scale), half true.
- Exposure. 64 statements, each assigned to a topic. They were not rated for truth, so as not to prompt scepticism.
- Tests: immediately, after 1 day, after 1 week (7–8 days) and after 1 month (28–32 days).
- Each test had 16 repeated and 16 new statements.
- Each repeated statement was tested only once.
- Analysis. Cumulative-link (ordinal) mixed models with participants and items random. Preregistered hypotheses: H1, a main effect of repetition; H2, a repetition × interval interaction.
Main results (verified)
-
H1 supported: β = 0.57 (SE .04), χ²(1) = 171.88. Repeated 4.51 (SD 1.45) against new 4.13 (SD 1.34).
-
H2 supported: interaction χ²(3) = 121.15. The effect shrinks with delay:
Delay Repeated − new (1–7 scale) Log-odds Share of the immediate effect (my computation) Immediately 0.68 (4.80 against 4.12) 1.04 1 1 day 0.39 0.56 0.57 1 week 0.27 0.37 0.40 1 month 0.14 0.20 0.21 All four are significant after Holm correction.
-
People showing the effect: 85.2% overall (483 of 567); by delay, 75%, 72%, 71% and 61%.
-
Age (exploratory). No relation to the overall effect (r_s = .06), but the trajectory differs:
- 50-year-olds: 0.76 immediately, 0.11 at one month;
- 25-year-olds: 0.59 and 0.18.
-
Attention (category accuracy .912) explained 0.53% of the variance.
-
Cited, not read in the original:
- Dechêne 2010 found no moderation by delay across studies: within-item d .25 (same day), .44 (within a week), .44 (longer); between-item .48, .43, .48.
- Brown and Nix 1996 (Experiment 2): an effect for false statements at one week that dissipated at one and three months.
- Nadarevic and Erdfelder 2014: no effect at 10 minutes, d_z .54 at one week.
- Silva et al. 2017: d_z 1.34 after minutes, .761 after a week.
Limits
- One exposure and one repetition per statement.
- Trivia with no strongly held beliefs.
- Truth was not rated at exposure; the authors note this may matter.
- WEIRD sample; data collected in lockdown; possibly conscientious Prolific workers.
- Inconsistencies found:
- The abstract gives χ²(1) = 169.91 and means of 4.52 and 4.14; the results give 171.88, 4.51 and 4.13.
- The discussion gives the immediate difference as 0.67; Table 5 gives 0.68.
What it means for Kurisutina (inference)
- The best direct reference in hand for how a change without reasons decays. After a single repetition:
- about half of the boost is gone within a day;
- about a fifth remains at one month;
- it is still detectable at one month in 61% of people.
- For M1's c:
- a non-zero carry-over of an unsupported repeated claim is human;
- a carry-over that never decays is not.
- Interpolating linearly in log time puts about 0.29 of the immediate effect left at 16 days, M1's horizon (my computation).
- The age pattern is a candidate setting for B6: older people show a larger immediate effect and faster loss.