Writing notes by hand: smaller and stranger than the internet says

silicon-canals piece walking back the popular claim that handwriting notes beats typing. Its own summary of the position — the advantage is “smaller and stranger than the internet says” — is a fair description of the three studies it assembles. Arrived via Telegram 2026-08-12 and routed here because the subject is learning and memory measured experimentally, which is this spoke’s declared territory, and because the piece is structurally a replication-crisis story.

The three studies it reports

A 2024 meta-analysis in Educational Psychology Review, combining 24 studies across 21 articles of college students taking lecture notes by hand or by keyboard, then reviewing and being tested. Handwriting wins on test performance at Hedges’ g = 0.248 — the article’s own reading is “a real result, but a small one.” Typing wins, and by much more, on how much material gets recorded: Hedges’ g = 0.919.

A 2021 direct replication of the Mueller & Oppenheimer (2014) study that started the whole argument. 74 laptop users and 68 longhand users took notes during lectures. “On the immediate quiz, however, neither group performed better.” The article reports two secondary findings that point the same way: more words recorded helped both groups, and verbatim copying hurt performance in both.

A 2026 EEG study in Frontiers in Education: 33 South Korean undergraduates, three ten-minute video lessons, notes taken with ink pens, digital pens, or keyboards. “The immediate tests found no statistically significant performance difference.”

What the assembled evidence actually says

The popular claim is that paper does something special to memory. What these three studies support is narrower and more interesting: the mental work done while taking the note matters more than the instrument. Handwriting’s small edge is attributed to a constraint — you cannot write fast enough to transcribe, so you must compress and rephrase, and that processing is what shows up on the test. Typing’s much larger edge on volume is the mirror image of the same fact.

Read that way the two effect sizes are not in tension, they are the same mechanism seen twice. It also explains the null results: when a study measures immediate recall rather than the kind of test that rewards reorganized material, the constraint has nothing to pay off against.

Note which direction the correction runs. The finding did not vanish; it shrank and changed shape. g = 0.248 is an effect that survives testing and does not survive being put in a headline.

Evidence quality, stated plainly

T3, and the reason is the one already recorded on silicon-canals: editorial review is review of the writing, not of the science. Specifically —

• The article names no researchers and no institution for the 2024 meta-analysis, the study carrying its headline number. It gives venue and year and nothing else.
• The 2021 replication is reported without authors or venue.
• Only the 2026 EEG study is fully enough described (n, country, design, venue) to be looked up.

This is the second time this publisher has given this corpus a study it cannot open. replication-crisis already records the first: dunning-kruger-misread reported a successful large replication while naming neither its authors nor its title. The pattern is now two for two, and it is a property of the outlet rather than of any one piece.

Related

note-taking-and-learning · replication-crisis · silicon-canals · dunning-kruger-misread · why-writing-feels-hard · dysgraphia · synthesis