Testing Effect: Why Taking a Quiz Beats Rereading the Notes

Pulling a fact out of memory changes the memory. Rereading leaves the fact where it was and makes the page feel familiar, which is pleasant and misleading: familiarity on the page is not the same as the ability to produce the fact with the page closed. A test demands the production, and the effort of retrieving a half-formed memory strengthens the route to it, so that the next retrieval is easier and the memory lasts longer. The effect appears even when the test gives no feedback, so it is not a matter of learning the answers from the test; it is the act of retrieval that does the work.

Two features of the effect make it easy to miss. The first is timing. Right after learning, rereading often produces better recall than testing, because the material is fresh and a second reading tops it up; the advantage of testing shows up days later, when the reread material has faded and the retrieved material has not. Anyone who judges a study method by recall at the end of the session will pick the wrong one. The second is feeling. A reread passage flows; a test exposes what you do not know, which is uncomfortable. Roediger and Karpicke's students rated repeated reading as the method that would serve them best in a week, and it served them worst.

Students act on that feeling. When Karpicke, Butler and Roediger surveyed 177 college students in 2009 about how they studied, 84 percent listed rereading among their strategies and 55 percent ranked it first; 11 percent mentioned testing themselves, and 1 percent ranked it first. The method with the strongest evidence behind it was close to the bottom of the list.

The effect has limits that matter in practice, which Roediger and Karpicke set out in a 2006 review of the literature. A retrieval attempt that fails completely teaches little unless feedback follows; testing on material never learned is no substitute for learning it. The benefit is largest when the test requires producing the answer, as in free recall or short answer, rather than recognizing it among options, and when retrieval takes some effort. And the gain is clearest for the material retrieved. Whether it carries over to untested material depends on how closely that material is related: the review reports cases in which retrieving some facts helped memory for related untested facts, and cases in which it hurt them, and a quiz on chapter three does nothing for an unrelated chapter four.

Henry Roediger and Jeffrey Karpicke published "Test-enhanced learning: Taking memory tests improves long-term retention" in Psychological Science in 2006. The effect itself was much older; their contribution was to show it cleanly with educational material and to measure what students believed about it.

In the first experiment, 120 Washington University undergraduates read two short prose passages from a test-preparation book, one about the sun and one about sea otters, each of about 260 words and scored in 30 idea units. Each student studied one passage twice, in two seven-minute periods, and studied the other once and then took a seven-minute free recall test on it, writing down as much as they could. No feedback was given. A final recall test followed five minutes, two days or one week later. After five minutes, the twice-studied passage was recalled better, 81 percent of idea units against 75 percent. After two days the order reversed, 54 percent for the studied passage against 68 percent for the tested one, and after a week it was 42 percent against 56 percent. The tested passage was recalled as well after a week as the studied passage had been after two days.

The second experiment, with 180 students, compared three schedules over four five-minute periods: study four times, study three times and test once, or study once and test three times. The students then rated how well they expected to remember the passage in a week. Those who had studied four times were the most confident. On a final test five minutes later they did recall the most, 83 percent against 78 and 71. A week later the order was reversed: 40 percent for the study-only group, 56 percent for the group with one test, and 61 percent for the group with three tests, who had read the passage an average of 3.4 times against 14.2 for the study-only group.

The authors drew the practical conclusion directly. Testing is usually treated as a way of measuring learning; it is also one of the most effective ways of producing it, and the students' own predictions ran the opposite way from their results.

The effect applies wherever something has to be remembered later, and the choice between rereading and recalling comes up more often than it seems. The rows below are illustrations, not findings from the studies cited on this page.

The testing effect is an advantage rather than an error, so the counter-measures are for the bias that hides it: the preference for the method that feels better.

• Test before you reread. Close the material and produce what you can; only then reopen it. The reopening becomes feedback instead of a third pass.
• Make the test demand production. Free recall or written answers beat recognizing the right option, because they require the retrieval that does the work.
• Judge a method by delayed results, not by how the session felt. The students who felt most confident after four readings recalled the least a week later.
• Space the tests out. Retrieval after a gap is harder and more useful than retrieval straight after study, which is the spacing effect working alongside this one.
• Pair failed retrievals with feedback. A blank is not a lesson; a blank followed by the answer is.
• If you teach, give frequent low-stakes quizzes and treat them as learning rather than measurement.

The testing effect is one of several findings about how the manner of learning shapes what is kept, and it is often blended with its neighbors.

• It is not the spacing effect, which is about when study happens: two sessions with a gap beat one session of the same length. The testing effect is about what happens in a session, retrieval rather than rereading. The two combine, and the best schedules use both.
• It is not the generation effect, in which material you produce yourself, such as a word completed from a fragment, is remembered better than material you read. That effect is about the first encoding; the testing effect is about retrieving something already learned.
• It is not the levels-of-processing effect, where material processed for meaning is remembered better than material processed for its surface. Testing works whether or not the material was processed deeply, and it works on top of that effect rather than through it.