Retrieval Practice in the Classroom: What the Studies Show and How to Use Them

Retrieval practice works in real classrooms: quizzing students on material they have already studied raises achievement by about half a standard deviation. A 2021 meta-analysis in Psychological Bulletin pooled 222 studies and 48,478 students and found an average gain of g = 0.50, roughly the gap between a student at the 50th percentile and one at the 69th.

That meta-analysis also says a good deal about what makes a quiz pay off: feedback after it, repetition across a unit, and a format that resembles the exam. It is thinner on finance classes, which is worth knowing before a unit is built around quizzing.

Assessment that can stay low-stakes

Yang's team found that quizzes helped whether or not the score counted toward a grade, so a quiz does not need to count to help. In the Rapunzl curriculum, assessment grades integrate with most LMS and SSO providers, including Clever, ClassLink and Google Classroom.

Frequent checking adds work for a teacher. It helps when grades flow into the systems a school already runs.

No study here tested a finance curriculum's assessments, and the limits section covers how thin the finance evidence is.

Does quizzing students raise achievement?

Yes. Yang and colleagues note that hundreds of studies over the last century showed the testing effect, and their meta-analysis asks whether it holds in classrooms. The table puts the four sources side by side.

Yang and colleagues also report what the quizzes were compared against. Quizzing beat no activity or filler tasks (g = 0.61) and beat restudying (g = 0.33, about the 63rd percentile).

The gains showed up beyond memorized facts. The meta-analysis found benefits for factual learning (g = 0.52), conceptual learning (g = 0.64) and applying knowledge to solve problems (g = 0.45). Roediger and Karpicke, reviewing the field in 2006, put the broader claim cautiously: frequent classroom testing may boost achievement at all levels of education.

Does it work for middle and high school students?

McDermott's team ran quizzes in seventh-grade science and high school history. Students took intermittent short-answer or multiple-choice quizzes with correct-answer feedback on some material, while other material got equal coverage in class without being quizzed. On unit exams and an end-of-semester exam, the quizzed material came out ahead. A restudy condition helped less than quizzing with feedback, so rereading alone is unlikely to explain the benefit in that study.

Carpenter and colleagues followed 8th graders through U.S. history facts. Reviews came 1 week or 16 weeks after the course ended, either by a quiz with feedback or by restudying. Each group took a final test 36 weeks after its review, about nine months later. Facts reviewed through testing were retained significantly better than restudied ones and nearly twice as well as unreviewed ones. The best retention was for facts quizzed at 16 weeks, though the difference from the 1-week group was not significant. The authors note the gains were numerically small because of floor effects.

Yang's meta-analysis breaks the effect out by level: elementary g = 0.33, middle school g = 0.60, high school g = 0.66 and university g = 0.49. All four were statistically significant, but the differences between levels were not, so the high school figure is not evidence that high school students gain more. It rests on 49 effect sizes.

What makes a classroom quiz work better?

The moderator analyses in Yang et al. give teachers a short list of things worth getting right.

• Feedback helps. Quizzes followed by corrective feedback produced g = 0.54, against g = 0.37 without it.
• More quizzes on the same content help. Gains rose from g = 0.44 for one quiz to 0.60 for two and 0.64 for three or more.
• Format matters, and every format helped. Matching (g = 0.91), fill-in-the-blank (0.77), short answer (0.64) and multiple choice (0.57) all produced gains, and free recall, the hardest format, produced the smallest (0.24). Recall and recognition formats gave the same gain (about 0.52 each), so multiple choice is a sound choice. A quiz in the same format as the exam did better (g = 0.53) than a mismatched one (g = 0.40).
• Timing counts. Post-class quizzes (g = 0.54) outperformed pre-class quizzes given before the material was taught (g = 0.19), although both still helped.
• Stakes showed no reliable difference. High-stakes quizzes produced g = 0.44 and low-stakes quizzes g = 0.48, which was not a significant gap. The authors also note that higher stakes may induce test anxiety.

What does the research not show about retrieval practice?

Finance classes are a thin slice of the data. In Yang's subject breakdown, accounting, business, economics and finance had only 16 effect sizes and a lower average than mathematics and statistics, g = 0.31, with a confidence interval running from 0.00 to 0.62. Mathematics and statistics did better (27 effect sizes, g = 0.43). Subject did not significantly change the overall effect, so a difference has not been shown, and it remains open whether one exists. Nobody has measured the full size of the benefit in a high school personal finance course.

High school is also a small share of the evidence. It contributes 49 of the 573 effect sizes in the analysis, with university classes supplying 335.

Quizzing is not shown to beat every other strategy. Against other elaborative methods such as concept mapping or summarizing, quizzing showed no reliable advantage (g = 0.10, not statistically significant). The case is strongest against doing nothing or rereading.

Design affects the number. Studies that compared quizzed and unquizzed material within the same students found bigger effects (g = 0.67) than studies comparing separate groups (g = 0.42), so the 0.50 average blends both. In Yang's data, the benefit also reached content that was not directly quizzed, but was smaller there (g = 0.32 against 0.51). Carpenter's classroom study had floor effects that kept absolute gains small.

Roediger and Karpicke acknowledge that testing can have negative consequences in some circumstances, though they describe them as often small. Yang's team judged the risk of publication bias in this field to be small.

How do you build retrieval practice into a high school finance unit?

1. Open class with questions on earlier lessons, notes closed. Yesterday's topic is fine; so is last week's. A few questions is a manageable place to start, and that is practical judgment: the studies did not test how long a quiz should be.
2. Use whichever format is quick for you, ideally the one your exam uses (g = 0.53 when matched, 0.40 when not). Multiple choice is fine.
3. Reveal the answers right away and have students fix their own papers. Feedback was one of the clearest moderators.
4. Quiz on a topic after you teach it. Quizzes given after class (g = 0.54) beat pretests on new material (g = 0.19), and step 1's opener on old lessons counts as a post-class quiz.
5. Quiz the same ideas more than once. Gains rose with the number of quizzes on the same content. Bringing a week-two topic back in week six is one way to do it; the meta-analysis did not test how far apart the repeats should be.
6. Include apply-it questions as well as vocabulary. "Which loan costs more over five years, and why?" tests the conceptual and application learning where the meta-analysis found gains.
7. Keep the stakes modest. The benefit did not depend on them.

For other classroom routines that fit a finance unit, see 5 tricks for teaching personal finance. Teaching dollar-cost averaging covers one concept that is easy to bring back in later quizzes.

Frequent, low-stakes checks in a finance unit are worth talking through with our team: request a Rapunzl demo and ask how it could fit your classroom.

By Clarissa Collins, Curriculum Designer at Rapunzl.