The testing effect

Careful observers noticed the effect long before anyone tested it. In 1620 the English philosopher Francis Bacon wrote that reading a text twenty times teaches it less well than reading it ten times while trying, now and then, to recite it, and checking the text whenever memory fails.

In 1890 the American psychologist William James made the same point. When you almost know a passage by heart, he wrote, "it pays better to wait and recollect by an effort from within, than to look at the book again." Memory, he said, is impressed better by active repetition than by passive repetition.

The first controlled experiments came in the early twentieth century, beginning with a study by Edwina Abbott published in 1909.

Why would rereading win after five minutes and lose after a week? The psychologists Elizabeth and Robert Bjork proposed that every memory has two separate strengths. Its retrieval strength is how easily it comes to mind right now. Its storage strength is how deeply it is learned.

Rereading pushes retrieval strength up fast. The page looks familiar and the answers feel close at hand, so a test a few minutes later goes well. But that ease soon fades, and rereading adds little storage strength. Pulling an answer out yourself is slower and less comfortable, yet it builds storage strength, which is what remains once the easy feeling has gone.

So a smooth study session can mislead you. How well you perform while studying is unreliable evidence of how much you are learning.

If recalling builds memory, does it matter how hard the recall is? It does. Retrievals that take real effort but succeed strengthen a memory more than easy ones. In the Bjorks' terms, the effort builds storage strength.

In one experiment, people practised recalling words from cues that showed only some of each word's letters. The fewer letters they were shown, the more challenging the recall, and the better they remembered the words later. A generous hint makes practice easier and does less good.

Robert Bjork called this kind of obstacle a desirable difficulty: something that slows learning down in the moment but makes it last. The difficulty has to be one you can overcome, though. A question so hard that you retrieve nothing, and never learn the answer, teaches very little.

Not every test asks for the same work. A multiple-choice question shows you the answer among a few options, so you only need to recognise it. A short-answer question offers nothing to choose from, so you have to produce the answer yourself.

Producing an answer takes more effort, and in laboratory studies it usually does more for memory: one large review found bigger benefits from recall tests than from recognition tests. In classrooms the gap is less clear.

Multiple choice carries a second risk. Its wrong options are written to look right, and a learner who picks one can go on believing it later. That risk is much smaller when learners are shown the correct answers afterwards, which is where feedback comes in.

A test teaches most when you find out the right answers afterwards. This is feedback, and it matters a great deal. One large review found the benefit of practice testing nearly twice as large in studies where feedback followed the test.

Feedback matters most for mistakes. A wrong answer you produced can feel familiar the next time and slip back in. Seeing the correct answer turns the error into a chance to learn.

Surprisingly, the easiest mistakes to fix are the ones you were sure about. When people confidently give a wrong answer to a general-knowledge question and are then shown the right one, they are more likely to remember the correction than when their wrong answer was only a guess. The surprise seems to grab attention. Psychologists call this the hypercorrection effect.

Can a test help before you have studied anything? In a set of experiments, people answered questions about a passage before reading it. Later they remembered those answers better than people who had simply spent the extra time reading, even for the questions they had first got wrong.

This is the pretesting effect. A failed attempt is not wasted, provided the correct answer arrives afterwards. The attempt seems to prepare you to notice the answer and hold on to it when it appears.

The benefit is narrow, though. A review of the research found that questions asked in advance improved learning of the information they ask about, but gave no general boost to the rest of the material. So the best questions to ask beforehand point straight at the most important ideas.

Testing does more than strengthen what you have just learned. It can also help you learn what comes next. When people study several lists or sections in a row, each new one is often learned a little worse than the last, because the earlier lists interfere with it. This build-up is called proactive interference.

In a series of experiments, people who took a short test after each list learned the later lists better than people who kept studying without tests. The tests seemed to stop each list from blurring into the next. Researchers call this benefit the forward testing effect.

Short quizzes placed between parts of a recorded lecture worked in a similar way: viewers who were quizzed let their minds wander less, took more notes and learned more.

If testing works so well, why do most students reread instead? When university students are surveyed about how they study, rereading notes and textbooks comes out far ahead, and far fewer practise recall.

The reason is that rereading feels like learning. On a second pass the page is familiar and the ideas seem obvious, and it is easy to mistake that familiarity for being able to recall the material. Psychologists call this mistake an illusion of competence. Testing yourself feels worse, because it exposes every gap.

Students' own predictions show the illusion at work. In one experiment, students expected that drawing a concept map of a science text, with the text in front of them, would help more than practising recall. On a test a week later the recall group did better, even when the test asked them to draw concept maps.

Much of this research began with word lists in laboratories. Does the effect survive real classrooms, with real courses and exams? A 2021 review pooled 222 classroom studies involving more than 48,000 students. On later exams, students who had been quizzed on course material did better than students who learned the same material in other ways, by a medium-sized margin.

That changes what a test is for. Schools usually treat a test as assessment: a way to measure what students know once the teaching is done. The research suggests that frequent quizzes also belong inside the teaching itself, as a tool for teaching and not only for measuring.

A common worry is that testing only trains people to parrot facts. The evidence is more encouraging, with a condition attached.

Practice tests can help you use what you learned on new questions you never practised, a benefit called transfer. In the concept-map experiment, recall practice helped even on questions that asked students to draw inferences the text never stated outright. A review of many transfer studies found real but modest benefits overall, and larger ones when practice went beyond isolated facts: explaining an idea, answering questions about how its parts relate, or getting feedback that explains why an answer is right.

So the kind of question matters. Practice that asks only for a word transfers less than practice that asks why.