C7 Handwritten notes
narrowed 3 attacks: 3 narrowed · serves Q1, Q2 · last change 21 October 2026
Students who take lecture notes by hand score higher than students who type them on a laptop on conceptual questions in a test taken at least a day after the lecture, without the notes in front of them, when both groups get the same note-taking instructions — but not consistently: in a randomised classroom trial in which about 230 university economics students each took notes once by hand and once on a laptop, scores on graph and equation questions in a quiz two days later did not differ significantly (about 0.06 standard deviations lower after laptop notes; few questions, not preregistered; Artz et al. 2020), and in a laboratory replication that used the original study's lectures and conceptual questions, about 60 university students per note-taking group scored no differently two days later, whether they took the test without reviewing their notes or after reviewing them for seven minutes (scores near the floor in the first experiment; Morehead, Dunlosky & Rawson 2019).
In practice
When the aim is to understand the ideas, notes by hand may help more than typed notes, especially if typing would turn into word-for-word transcription; but the advantage has not held reliably in replications, so don't rely on the medium alone.
Go deeper
Where it holds
Main limits
- For conceptual understanding tested at least a day later without the notes; not shown for remembering facts (Horbury & Edmonds 2021 found no difference there).
- Not robust across settings: a preregistered replication with university students, in sessions with distractions and a quiz right after the lecture, found no difference (Urry et al. 2021); the original co-author reads this as a failure to generalise beyond distraction-free settings (Oppenheimer & Ellefson 2024).
- Another direct replication, with the original lectures and questions, found no difference on conceptual questions two days later, whether or not students first reviewed their notes for seven minutes (Morehead, Dunlosky & Rawson 2019; scores near the floor in the first experiment).
- In a randomised classroom trial where about 230 economics students each took notes both ways, graph and equation questions in a quiz two days later showed no significant difference (Artz et al. 2020; few questions, not preregistered).
All limits (16)
Inside the circle: the core of the claim. Outside: every bound and carve-out in the claim file, numbered, with the attack that forced or added it where the record shows one.
forced by a narrowing attackin the original wordingwritten at creation, or added without a narrowing
Warm cards in the list were added by later attacks; plain cards were in the original draft or written at creation.
Reworded 2026-10-16 (9 consecutive fruitless searches): the claim now names the delayed conceptual test that the studies in the library differ on.
Added on 16 October 2026 when the claim was reworded (not an attack)
Status stays weakened because the evidence is mixed: Horbury & Edmonds 2021 found a handwritten advantage on conceptual questions after one week; Engelen et al. 2025 found a small pen advantage on an immediate test that was gone a week later; Voyer, Ronis & Byers 2022 (meta-analysis, abstract only) found no average effect of method; Urry et al. 2021 tested immediately and so does not directly test the reworded claim.
Forced by attack #1 · Urry et al. 2021 · 7 October 2026
Scope: note-taking for conceptual understanding, tested without the notes, with the same note-taking instructions in both conditions.
Written when the claim was created (6 October 2026); not forced by an attack
Not about fact recall, handwriting skill or composition.
Written when the claim was created (6 October 2026); not forced by an attack
Evidence in the project: Mueller & Oppenheimer 2014 (three studies with university students: laptop note-takers did worse on conceptual questions, linked to verbatim transcription); Horbury & Edmonds 2021 (26 boys aged 10–11, per the verified abstract: no difference on factual questions, handwritten notes better for conceptual understanding after one week, in a history and a biology lesson); Flanigan et al. 2024 (meta-analysis of 24 studies with college students: taking and reviewing handwritten notes gave higher achievement, Hedges' g = 0.248, although typing produced more notes, per abstract); Morehead, Dunlosky & Rawson 2019 (direct replication with extension; see the consistency bound below for its two-day tests).
Written when the claim was created (6 October 2026); not forced by an attack
Wording changed later: Forced by attack #3 · Morehead, Dunlosky & Rawson 2019 · 21 October 2026.
Robustness bound (added 2026-10-07, nightly attack): Urry et al. 2021 (Psychological Science; preregistered direct replication of Mueller & Oppenheimer Study 1, N = 142 university students, five TED talks, quiz after a distractor without studying the notes; full text read in the authors' 2019 preprint, published abstract checked): conceptual items g = −0.13 (95 % CI −0.45 to 0.20), favouring the laptop slightly, equivalent to zero within ±0.49 (p = .016); factual items g = −0.08, equivalent to zero; laptop notes longer and more verbatim.
Forced by attack #1 · Urry et al. 2021 · 7 October 2026
Its sessions were held at various times outside class and were subject to distractions and errors; the original co-author argues it is a failure to generalise to such conditions rather than a failure to replicate, and that meta-analyses of mostly controlled studies (Allen et al. 2020, as reported; Flanigan et al. 2024) find a longhand advantage (Oppenheimer & Ellefson 2024).
Forced by attack #1 · Urry et al. 2021 · 7 October 2026
Brain-imaging and letter-learning studies in young children (Askvik, van der Weel & van der Meer 2020; James & Engelhardt 2012; Longcamp et al. 2005) and composition in grades 2–6 (Berninger et al. 2009) concern handwriting in general, not note-taking outcomes.
Written when the claim was created (6 October 2026); not forced by an attack
Consistency bound (added 2026-10-17, nightly attack): Artz, Johnson, Robson & Taengnoi 2020 (The Journal of Economic Education 51(2), 103–115; full text read): within-subject randomised classroom trial, 230 principles-of-microeconomics students in five sections, assigned by odd/even student ID to take notes by laptop or on paper for one stand-alone lecture and the other medium for a second lecture a month later; ten-question multiple-choice quiz two days later and two exam questions 2–3 weeks later.
Forced by attack #2 · Artz, Johnson, Robson & Taengnoi 2020 · 17 October 2026
Computer notes vs paper, within-student fixed effects: conceptual (graph and equation) subset −0.056 SD (t = −0.61), quiz −0.123 SD, exam −0.027 SD, none significant.
Forced by attack #2 · Artz, Johnson, Robson & Taengnoi 2020 · 17 October 2026
Not preregistered, few items, sample called relatively small by the authors, point estimates all slightly favour handwriting, review of notes before the quiz not described.
Forced by attack #2 · Artz, Johnson, Robson & Taengnoi 2020 · 17 October 2026
Consistency bound extended (2026-10-21, nightly attack): Morehead, Dunlosky & Rawson 2019 (Educational Psychology Review 31, 753–780; full text read in the published version posted by the first author): two randomised laboratory experiments with Kent State undergraduates, Mueller & Oppenheimer's TED talks and factual and conceptual-application short-answer questions, instructions to take notes "as they normally would in class", notes removed after the talk, laptops limited to Word.
Forced by attack #3 · Morehead, Dunlosky & Rawson 2019 · 21 October 2026
Experiment 1 (188 analysed, about 63 per group, sample size set for 0.90 power to detect the original conceptual effect, d = 0.76): conceptual questions two days later without any review, main effect of note-taking method F(2, 151) = 0.19, p = .83, η² < .01; the authors note overall performance was near the floor.
Forced by attack #3 · Morehead, Dunlosky & Rawson 2019 · 21 October 2026
Experiment 2 (61 longhand, 61 laptop, 63 eWriter, two talks): two days later, after seven minutes reviewing their own notes, conceptual F(2, 168) = 1.17, p = .31, laptop numerically highest (Fig. 4).
Forced by attack #3 · Morehead, Dunlosky & Rawson 2019 · 21 October 2026
No equivalence test and no preregistration reported; in Experiment 1's delayed-only group longhand was numerically ahead of laptop (Fig. 2); TED talks rather than course lectures.
Forced by attack #3 · Morehead, Dunlosky & Rawson 2019 · 21 October 2026
So the delayed conceptual advantage was not shown in either randomised test with the original materials.
Forced by attack #3 · Morehead, Dunlosky & Rawson 2019 · 21 October 2026
11 of 16 bounds and carve-outs were forced by a specific attack.
How it changed
The ledger records 3 attacks, but no earlier wording of C7 is on record, so the text cannot be replayed step by step. Only the current wording is shown.
Students who take lecture notes by hand score higher than students who type them on a laptop on conceptual questions in a test taken at least a day after the lecture, without the notes in front of them, when both groups get the same note-taking instructions — but not consistently: in a randomised classroom trial in which about 230 university economics students each took notes once by hand and once on a laptop, scores on graph and equation questions in a quiz two days later did not differ significantly (about 0.06 standard deviations lower after laptop notes; few questions, not preregistered; Artz et al. 2020), and in a laboratory replication that used the original study's lectures and conceptual questions, about 60 university students per note-taking group scored no differently two days later, whether they took the test without reviewing their notes or after reviewing them for seven minutes (scores near the floor in the first experiment; Morehead, Dunlosky & Rawson 2019).
What would change this
What would kill it: A preregistered or well-powered randomised study, or a meta-analysis of such studies, with the same note-taking instructions in both groups, in which laptop note-takers score as well as or better than longhand note-takers on conceptual questions tested at least a day after the lecture without the notes. Reworded 2026-10-16 from the earlier kill-condition (any preregistered replication with typing ≥ pen on conceptual outcomes), which did not specify the test delay that the studies differ on.
Attack record
Reworded on 16 October 2026 (a change of wording, not an attack, so it is not counted). After 9 nightly searches in a row found no study that could test the old wording: searches kept finding studies that differ mainly in test delay (immediate: Urry 2021, Engelen 2025; one week: Horbury & Edmonds 2021, Engelen 2025), which the old wording did not specify. The new wording names longhand vs laptop on delayed conceptual tests, keeps the claim's spirit and keeps status weakened. No new data added.
3 attacks: 3 narrowed. Counts of attacks, not a score.
Full attack list (3)
In a preregistered direct replication of Mueller & Oppenheimer's Study 1 with university students, the same note-taking instructions and a conceptual quiz taken without the notes, longhand note-takers did not outperform laptop note-takers (conceptual g = −0.13, statistically equivalent to zero), although laptop notes were longer and more verbatim, so C7's kill-condition is met.
Strongest reading and verdict reasoning
Steelmans to: C7 dies on "a preregistered replication in which typing is at least as good as handwriting on conceptual outcomes tested without the notes, in the same population, with the note-taking instructions held constant"; Urry et al. is exactly that (preregistered 2017-03-07, N = 142, random assignment, five TED talks, equivalence-tested), and it reproduces the verbatim-notes mechanism while the conceptual advantage disappears; with Morehead, Dunlosky & Rawson 2019 also finding no consistent difference, the unconditioned claim cannot stand as stated.
Why: Nightly-attacks 2026-10-07 Europe/Stockholm. Full text read = the authors' 2019 PsyArXiv preprint (osf.io/vqyw6); the published version (Psychological Science 32(3), 326–339) is closed and was not read beyond its PubMed abstract, which confirms N = 142, more words and more verbatim overlap in laptop notes, no longhand advantage on the quiz, and that results "do not support the idea that longhand note taking improves immediate learning via better encoding". Preprint numbers: conceptual g = −0.13 (95 % CI −0.45 to 0.20), different from the original g = 0.34 (p = .003), equivalent to 0 ± 0.49 (p = .016); proportions correct laptop .74 vs longhand .70; factual g = −0.08 (equivalent to zero, p = .008); laptop notes longer (g = −0.90) and more verbatim (g = −0.85). Arm sizes differ between preprint and published abstract and are not quoted; mini-meta numbers differ between versions (preprint k = 7, published "eight") and are not used. Kill withheld: (1) Oppenheimer & Ellefson 2024 (DOI 10.1037/stl0000427, accepted manuscript read) argue the replication differed in setting and fidelity — sessions "subject to distractions and errors", undergraduate experimenters not fully trained — so it is a failure to generalise rather than to replicate, and that controlled-setting meta-analyses (Allen et al. 2020, as reported; Flanigan et al. 2024, g = 0.248) find a longhand advantage; (2) immediate test only, no review of notes; Horbury & Edmonds 2021 found a conceptual advantage after one week in middle-school classrooms. Not a restatement: Morehead 2019 was only an open challenge in Bounds and the claim text was unconditioned; first ledger entry on C7. Claim and Bounds rewritten same turn; brief and the practical rule rewritten.
In a randomised within-subject classroom trial in which about 230 university economics students each took notes once by hand and once on a laptop, laptop note-takers did not score significantly lower than longhand note-takers on conceptual (graph and equation) questions in an in-class quiz two days after the lecture, so C7's unconditioned delayed conceptual advantage does not hold consistently.
Strongest reading and verdict reasoning
Steelmans to: C7's reworded claim is precisely about tests at least a day after the lecture with the same instructions; this is the only randomised classroom test with a delayed quiz and a separate conceptual subset found in the widened search, and its crossover design removes self-selection into laptop use, the usual confound in classroom studies that favour paper.
Why: Nightly-attacks 2026-10-17 Europe/Stockholm, widened-search slot. Full text read (journal version as posted on ResearchGate; publisher copy closed, Unpaywall: no OA copy; not stored). Fall 2016, five principles-of-microeconomics sections at a US Midwest regional university, 230 students assigned by odd/even student ID; a fourth professor gave two stand-alone 30-minute whiteboard lectures about a month apart and the groups switched medium between trials; identical ten-question multiple-choice quiz two days later plus two exam questions 2–3 weeks later; the "conceptual" subset (3–4 quiz items per trial) is questions that use or interpret equations or graphs. Within-student fixed effects, computer vs paper notes: conceptual subset −0.056 SD (t = −0.61), quiz −0.123 (t = −1.47), exam −0.027, total −0.119, none significant; unpaired t-tests on the conceptual subset p = .727 (trial 1) and p = .441 (trial 2). Kill withheld: not preregistered, the authors call the sample relatively small and the item count small, every point estimate slightly favours handwriting, there is no equivalence test, and the paper does not say whether or how students reviewed their notes before the quiz; its "conceptual" items are quantitative graph and equation questions, not Mueller & Oppenheimer's conceptual-application questions. Narrowed: the delayed conceptual advantage is not shown consistently. Not a restatement: since the 2026-10-16 rewording the claim text carried no consistency bound; Engelen et al. 2025 (Bounds) had no conceptual split and Voyer et al. 2022 is abstract-only in the project. Not a preprint. Claim and Bounds rewritten same turn; brief and the practical rule rewritten.
In a direct replication of Mueller & Oppenheimer with their own lectures and conceptual questions, powered to detect the original effect, students randomly assigned to take notes on a laptop or by hand, with the same instructions and the notes taken away, did not differ on conceptual questions two days later (F(2, 151) = 0.19, p = .83), nor in a second experiment after reviewing their notes for seven minutes two days later (F(2, 168) = 1.17, p = .31, laptop numerically highest), so C7's kill-condition is met.
Strongest reading and verdict reasoning
Steelmans to: C7 dies on "a preregistered or well-powered randomised study... with the same note-taking instructions in both groups, in which laptop note-takers score as well as or better than longhand note-takers on conceptual questions tested at least a day after the lecture without the notes"; this study was designed with 0.90 power for the original conceptual effect, randomised, used the claim's own paradigm and a two-day test, and found no difference in two experiments, the second numerically favouring the laptop.
Why: Nightly-attacks 2026-10-21 Europe/Stockholm (ordinary slot; a lead from the study-reading notes, checked first-hand tonight). Full text read: the published article as posted by the first author on ResearchGate (Springer copy closed; Unpaywall no OA copy; not stored); figures checked on the publisher's image server; OSF project osf.io/dyga5 holds materials and data, not the text. Experiment 1: 193 participants, 5 removed, about 63 per group (longhand 63, laptop 64, eWriter 61), sample size from a power analysis for 0.90 power to detect d = 0.76; delayed test two days later with no access to notes; conceptual: method F(2, 151) = 0.19, MSE = 0.03, p = .83, η² < .01; immediate conceptual longhand M = .20 vs laptop M = .17, d = 0.13, p = .31 (one-tailed); authors: overall performance "near floor". Experiment 2: 222 participants, 5 removed; longhand 61, laptop 61, eWriter 63, no-notes 32; two days later, after 7 minutes' review of their own notes, conceptual method F(2, 168) = 1.17, MSE = 0.05, p = .31 (Fig. 4: laptop numerically highest); immediate conceptual longhand M = .23 vs laptop M = .30, d = −0.35, p = .09. Their cumulating meta-analysis of the four direct replications (immediate tests): conceptual d = 0.17, 95% CI −0.07 to 0.40. Kill withheld: power was set for the original large effect (d = 0.76), not the small effects in later meta-analyses (Flanigan et al. 2024, g = 0.248); no equivalence test; no preregistration reported; Experiment 1 scores near the floor; in Experiment 1's delayed-only group (the cleanest two-day test) longhand was numerically ahead of laptop (Fig. 2); Experiment 2's delayed test measures notes plus review, not encoding alone; TED talks rather than course lectures. Not a restatement: the existing consistency bound rests on one classroom trial with quantitative graph and equation items (Artz et al. 2020); this is the claim's own paradigm and conceptual-application questions, and the Bounds had described Morehead et al. only as "did not differ consistently", without its two-day tests. The 2026-10-16 "do not re-emit Morehead 2019" note was written against the old, undelayed wording and before its full text was read. Also corrected in Bounds and Ages, same turn: Horbury & Edmonds 2021 studied 26 boys aged 10–11 (verified abstract), not "middle-school classrooms". Not a preprint. Claim and Bounds rewritten same turn; brief and the practical rule Limits updated; the practical rule headline unchanged (already says the evidence is mixed).
Evidence
13 linked sources: 3 supports, 8 limits, 2 challenges (first-pass draft, pending human review). supports: Flanigan et al. 2024, Horbury & Edmonds 2021, Mueller & Oppenheimer 2014 · limits: Ose Askvik et al. 2020, Berninger et al. 2009, James & Engelhardt 2012, Longcamp et al. 2005, Urry et al. 2021, Oppenheimer & Ellefson 2024, Engelen et al. 2025, Artz et al. 2020 · challenges: Morehead et al. 2019, Voyer et al. 2022