Retrieval practice — quizzing students on material instead of having them reread it — carries the strongest evidence rating in a 2007 federal practice guide from the Institute of Education Sciences. But a 2024 study in the International Journal of STEM Education found the average classroom gain was a modest 2%, not the dramatic boost teacher-training slides often promise.
What counts as retrieval practice?
The federal practice guide Organizing Instruction and Study to Improve Student Learning, published by the Institute of Education Sciences in September 2007 under panel chair Harold Pashler of UC San Diego, defines it plainly: closed-book quizzes that force students to pull an answer from memory, rather than recognize it on a page. Fill-in-the-blank and short-answer formats count. Multiple choice works too, as long as students answer without the material in front of them.
The guide draws a distinction teachers often blur. Rereading notes or highlighting a textbook is restudying. Answering a quiz question from memory, then getting corrective feedback, is retrieval. The guide’s panel rated the second practice far more effective than the first — and gave that finding its highest confidence rating.
Is the evidence actually strong?
By the guide’s own scale, yes — for the core technique. The panel rated “use quizzes to re-expose students to information” as Strong evidence, its top tier, citing the testing effect as robust across ages and subject matter and more effective than extra study time. Spacing that quizzing over several weeks earned a Moderate rating. Two adjacent tactics — asking pre-questions before new material, and using quizzes only to flag what needs restudy — were rated Low, with the guide noting the pre-question research came mostly from reading studies, not general classroom subjects.
That 2007 rating is nearly two decades old. It still anchors most professional-development material on the topic, but newer subject-specific studies complicate the picture.
| Recommendation | Evidence level | What it looks like in class |
|---|---|---|
| Use quizzes to re-expose students to material | Strong | Closed-book, short-answer quiz with corrective feedback afterward |
| Space learning over time | Moderate | Revisit a topic at least twice, several weeks apart |
| Ask pre-questions before new material | Low | A question about an upcoming reading, asked before assigning it |
| Use quizzes to flag what needs restudy | Low | A diagnostic quiz used only to identify weak spots, not to re-expose students to content |
The gap between Strong and Low matters for how a department should spend limited planning time. A closed-book quiz with feedback is worth building into every unit. A pre-question warm-up is worth trying, but a teacher who skips it isn’t ignoring settled science — the guide itself says the research base for that particular tactic is thin.
So why did a 2024 study find only a 2% gain?
Researchers Campbell Bego, Keith Lyle, Patricia Ralston and colleagues tracked 578 students across nine introductory STEM courses — biology, chemistry, engineering, math, physics and psychology — running five biweekly quizzes with content either massed on one quiz or spaced across three. The study, published in the International Journal of STEM Education in February 2024, used a within-subjects design built on single-paper meta-analyses, a method designed to catch effects a single classroom’s data might miss.
Across eight of the nine courses, spacing showed no significant benefit on its own. Only after adding calculus data did the pooled result turn statistically significant — a 2.06% improvement, with a 95% confidence interval of 0.16% to 3.97%. The authors flagged high variability between courses (I-squared = 89.2%), writing that spacing effects were clear in mathematics but inconsistent everywhere else they measured. That is a real, measured effect — just a far smaller one than the phrase “science of learning” tends to imply on a workshop slide.
Does it work the same in math as everywhere else?
Not entirely, according to a meta-analysis by Ewan Murray, Aidan Horner and Silke Göbel published in Educational Psychology Review in July 2025. Pooling 27 studies on spaced versus massed practice in mathematics, the authors found a small-to-medium effect (g = 0.28) favoring spacing. But when they isolated the testing effect specifically — quizzing versus simply restudying — across seven studies, the effect size dropped to 0.18, with a confidence interval that crossed zero. The authors concluded that, for math specifically, “the current literature does not provide conclusive evidence for a consistent effect of retrieval practice,” citing too few classroom-based math studies to say more.
Put together with the 2024 STEM finding, the pattern is consistent: spacing holds up better than pure retrieval testing, and math is the subject with the clearest data — not because retrieval practice fails elsewhere, but because fewer well-designed classroom studies exist outside it.
Neither team of researchers argued teachers should drop the technique. Both flagged the same underlying problem: laboratory studies on retrieval practice run in tightly controlled settings with isolated word lists or flashcards, while classroom studies have to survive real interruptions — absences, uneven prior knowledge, a curriculum that doesn’t leave room for a clean five-quiz design. That gap between lab effect sizes and classroom effect sizes is precisely what a 2007 guide, built mostly on lab data, couldn’t fully anticipate.
Does frequent quizzing stress students out?
That is the objection teachers raise most, and the available research runs the other way. A resource page from the UK training organization Evidence Based Education, summarizing 2014 research by cognitive scientist Pooja Agarwal and colleagues, states that classroom-based retrieval practice reduced exam anxiety and gave students a confidence boost rather than adding pressure. The same resource, citing a 2013 review by John Dunlosky and colleagues, notes that popular study habits like rereading and highlighting do not produce the same long-term retention as retrieval or spacing — part of why cognitive scientists have pushed the technique for over a decade despite the modest classroom-scale effect sizes turning up in newer studies.
How should a teacher actually build this in?
- Make it low-stakes. The federal guide’s strongest-rated recommendation describes closed-book quizzes with corrective feedback — not graded high-stakes tests. The retrieval act is what matters, not the grade attached to it.
- Space repeat exposure by weeks, not days. The guide’s Moderate-rated recommendation calls for revisiting material at least twice, separated by several weeks, rather than cramming repetition into a single unit.
- Use short-answer or fill-in-the-blank formats where possible. These force actual recall; recognition-based formats give a smaller retrieval boost, per the same guide.
- Treat pre-questions and diagnostic-only quizzes as unproven, not useless. Both carry the guide’s Low evidence rating — worth trying, not worth building a whole unit around.
- Expect a modest, subject-dependent gain, not a dramatic one. The 2024 STEM data put the measurable bump around 2% pooled across courses; treat larger claims with skepticism until a named study backs them.
The verdict, by classroom
For a single unit review, low-stakes closed-book quizzing is close to a free win: the federal guide’s strongest evidence rating, plus a mechanism — corrective feedback after active recall — that costs a teacher little beyond quiz-writing time. For year-long retention claims, the 2%–28% effect sizes turning up in 2024 and 2025 research are real but modest, enough to justify building spacing into a syllabus, not enough to promise a specific test-score jump to a principal or a parent.
For a math department deciding where to spend planning time, spacing has clearer support than pure retrieval testing does; the sources reviewed here did not establish a reliable math-specific testing-effect size, and neither study claimed one. For a department in biology, chemistry, engineering or physics, the 2024 STEM data is the most direct classroom evidence available, and it argues for treating retrieval practice as one useful, modest-effect tool among several — not the single lever that turns around a struggling unit.
What none of these sources establish: a guaranteed grade-point or test-score outcome for any individual class, or an effect size specific to elementary or middle-school students, none of whom appeared in the studies reviewed here.
For a related edtech news perspective, read British Royal music featured on Gossip Stone TV reality TV show by Debbie Wingham.