Acute stress just before or during a memory test lowers adults' recall of earlier-learned material by about a fifth of a standard deviation compared with no stress, more for emotional material, but only in studies that left out women taking hormonal contraceptives; where they were included, the average effect was about zero.
Status
weakened
Last change
2026-11-04
Serves
Q1 (How do humans actually learn, and what helps them learn best?)
Ages
vuxen-livslangt, overgripande
Area
Emotion, stress and arousal
Relates to
C25 relates-to C26 (both about stress and emotion around tests)
Bounds
Laboratory stressors (for example public speaking or cold water) shortly before or during retrieval of episodic memories, mostly in young adults; not chronic stress, not stress during learning (Shields et al. find that pattern differs), and not school exams directly. Evidence on file: Shields, Sazma, McCullough & Yonelinas 2017 (Psychological Bulletin, DOI 10.1037/bul0000100; full text): stress just before or during retrieval impaired memory, g = −.215 (95% CI −.346 to −.085; 31 studies, N = 1,410), larger for emotional than neutral material; Egger's test showed publication bias for these retrieval effects (t(29) = −3.27, p = .003), and trim-and-fill estimated one missing study, after which the impairment was still significant (p = .002), so the true effect may be somewhat smaller than −.215 (corrected 2026-11-03). Population bound (weakened 2026-11-04, nightly attack; Shields et al. 2017 full text, §Retrieval and Figure 9): of the moderators tested (age, sex, contraceptive exclusion, menstrual-phase exclusion, stressor type, memory task, encoding–retrieval delay, time of day), only exclusion of women taking hormonal contraceptives moderated stress effects on retrieval (t(8.1) = −3.44, p = .009; all other ps > .185): studies that excluded them g+ = −.294 (t(23.2) = −4.14, p < .001); studies that did not g+ = .101 (t(5.2) = 1.15, p = .302; the authors call it negligible). Caveats: "did not exclude" also covers studies that did not report exclusion, and the not-excluded subset is small (df = 5.2). Consistent with this, a registered-report experiment with 177 adults (82% women, about half of them on hormonal contraceptives; stress 10 minutes before the test) found no stress effect on recall or recognition (Marr, Sauerland, Otgaar, Quaedflieg & Hope 2021, Memory 29(3), 330–344, DOI 10.1080/09658211.2021.1893750; powered only for a medium effect). Klier & Buratto 2020 (Trends in Psychiatry and Psychotherapy, DOI 10.1590/2237-6089-2019-0077; abstract only): systematic review of 13 controlled studies (N = 962) of stress after learning and up to an hour before the test. Claim sentence shortened 2026-11-04 (Jonatan approved the short-claims rule): the subgroup sizes (excluded g+ = −.294; not excluded g+ = .101, ns) are kept above; meaning unchanged.
What would kill this
A meta-analysis or large preregistered study of controlled human experiments finds no impairment, or a reliable improvement, when acute stress comes shortly before or during retrieval. Promoted from K56 on 2026-10-29 in the bold expansion Jonatan approved after four empty attack nights (26–29 Oct): new sharp claims in thin areas, each naming a comparison, a rough size or direction and a population.
Tutor consequence
none yet
Attack record
1 attack: 1 narrowed
2026-11-04 narrowed
Attack
C26 states a general adult effect (about −0.2 SD at retrieval), but in the very meta-analysis it rests on, the only significant moderator of retrieval effects was whether studies excluded women taking hormonal contraceptives: excluded g+ = −.294 (p < .001), not excluded g+ = .101 (p = .302, "negligible"); a registered-report experiment with about half the women on contraceptives also found no impairment.
Target
claim-text
Source
Shields, Sazma, McCullough & Yonelinas 2017 | Psychological Bulletin 143(6), 636–675, DOI 10.1037/bul0000100 (full text, PMC5436944, stored privately; Crossref-checked 2026-11-04); consistent with Marr, Sauerland, Otgaar, Quaedflieg & Hope 2021 | Memory 29(3), 330–344, DOI 10.1080/09658211.2021.1893750 (registered report, stored full text)
Steelmans to
C26 is worded for adults in general; Shields' own moderator analysis (§Retrieval, Figure 9) shows that the pooled −.215 is carried by studies that left out contraceptive users, and in studies that included them the average was close to zero and not significant. Queued by study mining 2026-11-04.
Why
Nightly-attacks run of 2026-11-04 Europe/Stockholm. Checked tonight in the stored full text: "only exclusion of contraceptives emerged as a moderator of stress effects on retrieval (all other ps > .185), t(8.1) = −3.44, p = .009"; "If a study … did not exclude women taking hormonal contraceptives, the effect of stress on retrieval was negligible, g+ = .101, t(5.2) = 1.15, p = .302"; excluded: g+ = −.294, t(23.2) = −4.14, p < .001; largest reliable effect (emotional items, contraceptive users excluded) g+ = −.388. Marr et al. 2021 (N = 177, 82% women, about half on contraceptives): no stress effect on recall or recognition (planned ps .20–.70), but powered only for a medium effect, so it is supporting context, not the basis. Steelman for the claim: "not excluded" also means "not reported", the subset has few degrees of freedom (df = 5.2), and the pooled estimate across all studies is still −.215 and significant; so this narrows the population, it does not kill. Not a duplicate (the 2026-11-03 entry was the publication-bias correction); not a restatement of an existing Bound (Bounds named young adults and lab stressors, not contraceptive use). Not a preprint. Kill-match check (the routing method): the new text concedes a null subgroup, but the kill line ("a meta-analysis or large preregistered study … finds no impairment, or a reliable improvement") still names a result the claim does not concede for studies of men and non-users; not a kill-mismatch.
Rewrite
claim-field+brief
Corrections to our notes
These fix a number or a fact in our notes. They are not attacks and do not count toward the floor of 3 survivals.
— C26's limits report Shields et al. 2017's g = −.215 for stress at retrieval without the paper's own publication-bias finding for that effect.
— Wording change only, not an attack: the claim sentence was reworded without new evidence.
Full text: preregistered, 177 adults; stress just before retrieval did not significantly lower recall or recognition; powered only for a medium effect. Lead, not applied.
Weakened C26 2026-11-04: exclusion of hormonal-contraceptive users the only significant moderator of retrieval effects (t(8.1) = −3.44, p = .009); excluded g+ = −.294 (p < .001), not excluded g+ = .101 (p = .302, negligible). Consistent with Marr et al. 2021 registered-report null (about half the women on contraceptives).
Review state: first-pass draft, pending human review.