Study evidence check: what the learning-science papers actually say, quoted

Why this page exists

We cite learning-science research on our study pages, which means we can be wrong in a way readers cannot easily catch. So we checked: every citation, against the PubMed record, with the abstract read rather than the citation trusted. The check found real errors in our own pages, and correcting them is the reason this page can be specific rather than reassuring. What follows is the result, not a summary of the field.

What "verified" means here

For each claim below there is a PubMed identifier and a quotation copied from the abstract of that record. Not a paraphrase, and not a quotation carried over from another article that cited it. Second-hand quotation is how figures drift: each retelling rounds, simplifies, or attaches the number to a nearby claim, and after a few hops the sentence in circulation is not in any paper.

The two techniques with the strongest support

Dunlosky and colleagues, 2013 reviewed ten study techniques and rated them by how well their benefits generalise. Two came out on top: “Practice testing and distributed practice received high utility assessments because they benefit learners of different ages and abilities and have been shown to boost students' performance across many criterion tasks and even in educational contexts.” Testing yourself, and spreading review over time. That is the short list.

The five rated low utility, including two that feel productive

The same review is blunt about the rest: “Five techniques received a low utility assessment: summarization, highlighting, the keyword mnemonic, imagery use for text learning, and rereading.” Highlighting and rereading are the two most students report using. Low utility does not mean useless; the review says the conditions under which they help are narrow. It does mean they are a poor default.

What the spacing research says about intervals

The interval question has a specific answer and it is a ratio, not a fixed number of days. Cepeda, Vul, Rohrer, Wixted and Pashler, 2008 reports: “the optimal gap declined from about 20 to 40% of a 1-week test delay to about 5 to 10% of a 1-year test delay.” The gap scales with how far away the test is, and the proportion shrinks as the horizon grows.

A number that circulates and is not in the paper

A figure we had used ourselves, that review should sit at “10 to 20 percent of the horizon”, is attributed to that same 2008 paper. It is not in it. The paper's own range is the one quoted above, and it is two ranges rather than one because it depends on the test delay. We removed the figure from 25 of our pages. If you have seen it cited, the citation does not lead where it claims.

Two records that get cited as this paper and are not

While checking, we found two PubMed identifiers circulating in place of the real 2008 spacing paper. 19017207 is a microscopy paper on protein assembly in the neuronal porosome complex. 19086747 is the American Psychological Association's evidentiary review of zero-tolerance school policies. Neither is about spacing. The correct record is 19076480.

Restudying and testing are not the same lever

Karpicke and Roediger, writing in Science in 2008 put the two side by side and the result is unusually clean: “Repeated studying after learning had no effect on delayed recall, but repeated testing produced a large positive effect.” The same paper reports why students choose the weaker one: “students' predictions of their performance were uncorrelated with actual performance.” Your sense of what you know is not evidence about what you know.

How to check a citation yourself in about a minute

Take the identifier, open the PubMed record, and read the abstract for the sentence the claim rests on. If the sentence is not there, the claim needs the full text or it needs dropping. Check the record for an erratum line too: the longhand-versus-laptop note carries a 2018 correction, which is why we describe its mechanism rather than its effect size. This costs a minute and it is the whole method.

Last updated: 2026-08-21 · Written with AI assistance and reviewed before publishing.