Contribution
A preregistered, multisite direct replication of a central finding of terror management theory (TMT), designed to test a second question: does involving the original authors make a replication more likely to work? Labs were randomly assigned to run either an “Author Advised” protocol, reviewed by the original authors before data collection, or an “In House” protocol built from the published paper alone. The first question got a clear answer. The second did not, because the effect failed under both.
Key insight
The effect did not appear even where everything the theory’s authors asked for was done. The essay was made more extreme, a delay was added, the setting was made relaxed and anonymous, and the strictest exclusion rule kept only white, US-born participants who rated their American identity 7 or more out of 9. The pooled estimate stayed near zero. The authors’ conclusion (Discussion): “With these protocols, in the context of these labs and time in history, we find little support for this key finding of TMT.”
Assumptions
- The target is representative. Study 1 of Greenberg et al. (1994) was chosen with Pyszczynski and Solomon as theoretically central and likely to replicate. Its manipulation, writing about one’s own death, is the one 79.8% of TMT studies use, a figure the authors take from Burke et al. 2010 (Introduction).
- The dependent measure is worldview defense. Preferring a pro-US to an anti-US author, after a death prime, is read as defending one’s cultural worldview. The measure is the mean of three author ratings for the pro-US essay minus the mean of three for the anti-US essay (α = .89 for each), on 9-point scales.
- Two of the original five conditions. Only “subtle own death salient” (here called mortality salience) and “TV salient” (control) were run. The author ratings, not the essay ratings, were analysed, because they gave the strongest original effect.
- Two-tailed tests throughout. The preregistration did not say which; the authors say they always intended two-tailed tests, against Chatard, Hirschberger and Pyszczynski (2020), who argue for one-tailed (footnote 5).
Key results
- The original (Method). 59 US introductory psychology students across five conditions. Pro-minus-anti author difference: mortality salience 12.25, TV 1.64; t(53) = 4.87, p < .001, d = 1.34.
- The sample (Participants; Table 1).
- 21 labs at 18 universities collected 2,281 participants, in the US fall of 2016 and spring of 2017.
- One site stopped at 19 participants and was dropped (footnote 1). Four labs with fewer than 60 were excluded (158 participants). Data some In House labs collected before the preregistration were excluded (545). An earlier preprint had omitted the first exclusion, which Chatard et al. (2020) flagged; this version keeps to the preregistration (footnote 2).
- That left 17 labs and 1,578 participants, and 1,550 after incomplete responses were dropped: 699 in 7 Author Advised labs, 851 in 10 In House labs. 54% women, mean age 19.91 (SD 2.49), 61% White.
- The two protocols (Method; Table 2).
- Author Advised: always in person and on paper; a harsher anti-US essay; a longer filler sequence (a 23-item “personality inventory”, a 12-item need-for-structure scale before the prime; the 60-item PANAS-X and the 19-item Morningness-Eveningness Questionnaire after it); casual, relaxed experimenters; a covered box for completed packets.
- In House: all 12 such protocols were computer-administered, which the original authors had discouraged. Five had no filler task between the prime and the essays. None used the Author Advised fillers.
- Exclusion sets and power (Analysis Plan). Two stricter sets came
from the original authors, who did not agree among themselves. They
could be applied only to Author Advised data.
- Set 1, minimal: N = 1,550; 95% power for d = 0.18.
- Set 2, also only White and US-born: N = 1,229 (378 Author Advised); 95% power for d = 0.21.
- Set 3, also American identity 7 or more of 9: N = 1,076 (225 Author Advised); 95% power for d = 0.22.
- Question 1: the pooled effect (random effects).
- Set 1: g = 0.07, 95% CI [−0.03, 0.17], p = .187.
- Set 2: g = 0.09, [−0.03, 0.21], p = .135.
- Set 3: g = 0.09, [−0.04, 0.22], p = .193.
- No heterogeneity: Q(16) = 19.35, 21.72 and 18.57; p = .251, .152, .292.
- The preregistered three-level model failed to converge and was replaced by a univariate one (Deviations).
- Question 2: protocol as moderator (mixed effects).
- Grand means g = 0.07 (p = .182), 0.11 (p = .087), 0.13 (p = .117).
- Protocol: b = 0.01 (p = .842), 0.05 (p = .410), 0.07 (p = .391). Author Advised labs did not do better.
- Question 3: standardization. Neither protocol group showed between-lab variance; a random-effects model did not fit better than a fixed one (In House χ²(1) = 0.67, p = .41; Author Advised χ²(1) = 0.00, p = 1 under every set). The In House labs’ procedural variety did not show up as variable results.
- Per lab (Table 3, Set 1). Effects ran from g = −0.36 to 0.73. Only one lab was significant: Illinois, an In House lab, g = 0.73, p < .001. The largest Author Advised effect was g = 0.42 (BYU–Idaho, p = .06).
- Exploratory analyses (not preregistered).
- Author Advised labs alone: g = 0.08 (p = .292), 0.17 (p = .111), 0.19 (p = .200).
- The five labs whose leaders reported “a lot” of TMT knowledge: g = 0.02, 0.03, −0.03.
- Participants preferred the pro-US author overall (M = 5.68 against 4.98). The harsher essay worked as intended: 68% at Author Advised labs preferred the pro-US author, against 48% at In House labs (χ²(2) = 76.82).
- Restricted to Author Advised participants who preferred the pro-US author: g = 0.15 (p = .110), 0.13 (p = .306), 0.07 (p = .649).
- The researchers’ expectations. 23 researchers from 17 sites put the chance of success at 56% on average (range 20% to 95%). Ten hoped it would replicate, two hoped it would not.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Writing about one’s own death does not raise preference for a pro-US over an anti-US essay author among US students in 2016–17 by any effect larger than about g = 0.2 | strong for this paradigm, population and period: 17 labs, preregistered, narrow intervals under three exclusion rules | Question 1 |
| C2 | A protocol designed with the original authors produced no larger effect than protocols built from the paper | strong as a null on the moderator, but uninformative about expertise, since there was no effect to moderate | Question 2 |
| C3 | Labs that varied widely in procedure did not vary in result | moderate: absence of detectable heterogeneity, which low true effects make easy | Question 3 |
| C4 | The failure cannot be put down to participants not holding the pro-US worldview | moderate: the harsher essay shifted preferences, and restriction to patriotic participants or to those preferring the pro-US author left the effect non-significant | exploratory analyses |
| C5 | The original effect was a false positive, or the conditions for obtaining it are not understood or no longer exist | the authors’ own disjunction; the data do not choose between its parts | Abstract, Discussion |
Concepts
- mortality salience (MS): a manipulation that makes one’s own death salient, here two open-ended questions about the emotions one’s death arouses and what happens to one’s body as one dies and after.
- worldview defense: favouring those who uphold one’s cultural worldview and derogating those who challenge it; here, the gap between ratings of a pro-US and an anti-US author.
- Author Advised / In House: the replication protocol reviewed by the original authors, against one each lab built from the paper alone.
- exclusion sets: the three preregistered rules, two of them proposed by original authors to restrict the sample to those for whom the pro-US essay expresses their own worldview.
Connections
- Burke, Martens and Faucher 2010 (LIT-1687), the meta-analysis this note stands in for. The paper uses it twice: for the 79.8% share of TMT studies using this manipulation, and as a source on TMT and intergroup conflict. It gives none of its effect sizes. Chen et al. 2025, read alongside, does.
- Chen et al. 2025, the bias-corrected meta-analysis, describes this study as 21 labs and 2,220 participants with g = 0.03. Neither figure matches this version (2,281 collected, 1,550 analysed, g = 0.07). Chen et al. may be citing the earlier preprint.
- Chatard, Hirschberger and Pyszczynski (2020), the original authors’ critique, is cited for two points: the omitted exclusion, which this version corrects, and one-tailed tests, which it declines.
- Two other preregistered failures are named: Sætrevik and Sjåstad (2019 preprint) and Schindler, Reinhardt and Reinhard (2021).
- Castano et al. (LIT-1719) reminded participants of their death and measured in-group identification. It is not tested here; it depends on the same family of manipulations.
Bearing on the record
- The legacy question (QUESTION-301). Its “Death and the group”
strand rests on TMT: death reminders make people cling to a worldview
and a group that outlast them, which is symbolic immortality through a
self extended in time and across people. These are inferences; the
paper does not discuss the self or legacy.
- The flagship test of that strand failed. Writing about one’s death did not make US students favour a defender of their country, at any size the study could detect (g above about 0.2), with the original authors’ protocol or without it.
- The test is of national worldview defense, the group half of the strand. It does not test symbolic immortality directly, nor legacy motivation, generativity or beneficence toward future generations. The legacy experiments (LIT-1716, LIT-1760, LIT-1724) are unread. Wade-Benzoni et al. (LIT-1760), whose title names death, may rest on a death prime; if so, this failure is a reason to check its effect sizes when it is read.
- It does not refute TMT. The authors grant that one failure “does not overturn all of that prior work”. It removes the assumption that the core effect is robust; Chen et al. 2025 says what remains.
- The owner’s thesis (CLAIM-491). The thesis does not rest on TMT, so nothing in it changes. TMT offered a mechanism for enlarging the self under threat; that mechanism now lacks a dependable demonstration.
- The triviality objection (CLAIM-492). Worldview defense is measured by evaluative preference for a pro-group author. That is a measure of caring about the group’s standing, not of the self’s extent. Even a successful replication would not supply the independent criterion the objection asks for (an inference).
- A dated note should be added to QUESTION-301, jointly with Chen et al. 2025. Proposed text: “Many Labs 4 (NOTE-1322) found no effect of writing about one’s death on preference for a pro-US over an anti-US essay author: g = 0.07 (95% CI −0.03 to 0.17) across 17 labs and 1,550 students, with or without the original authors’ protocol. Chen et al. (NOTE-YYYY) re-ran the mortality-salience literature with bias corrections and put the average effect near r = .18, not the r = .35 of Burke et al., with very wide heterogeneity and low power. The ‘death and the group’ strand of this question therefore rests on a small, contested effect whose flagship paradigm does not replicate.” No note is needed on CLAIM-491.
- No THEORY filed.
- No ML instruction.
Limitations
- One paradigm, one country, one moment. US students during the 2016 presidential election and after. An original author predicted the timing would mask the effect; the authors tried to counter it but cannot rule it out (Discussion).
- Expert subtleties are hard to verify. One original author would endorse no replication he did not supervise. Advice on experimenter manner was filmed but not checked.
- The stricter exclusions cut Author Advised samples to 378 and 225, so the analyses closest to the original authors’ wishes are the least precise.
- Typos and inconsistencies.
- The Introduction gives 1,578 participants “after exclusions”; the abstract and analyses use 1,550.
- The anti-US author’s SD is printed as 4.98, the same as its mean.
- “t(1, 549)” for a t-test with 1,549 degrees of freedom.
- “direction replication” for “direct replication” (Discussion).
- “Rosenburg” for Rosenberg, and “Morning-Eveningness” (Table 2).
- Table 2 cites essays from “Greenberg et al. (1992)”, which is not in the references.
- The author contributions list “D. Dudley”, who is not in the author list.
- In the references, the Nosek and Errington (2020b) entry runs into the Open Science Collaboration entry, and its article number is printed “e30000691”.
Open questions
- Would the effect appear with a measure of symbolic immortality itself, such as identification with a group seen as lasting, rather than defense of a national worldview? The record’s Castano et al. reading would bear on it.
- Under the one-tailed tests Chatard et al. prefer, halving the reported p-values would put the mixed-model grand mean under Exclusion Set 2 (two-tailed p = .087) just under .05. That is the record’s arithmetic, not the paper’s, and it concerns one of three preregistered rules.