Retrieval practice: why testing yourself beats rereading

You read the chapter again. The second time it goes faster, the terms look familiar, and the diagrams make sense. It feels like learning.

Most of that feeling is fluency: the text is easier to process because you have seen it recently. Fluency is a poor signal of what you will be able to produce in an exam hall a week later, with the book closed.

This post covers what the research says about testing yourself instead of rereading, why it feels worse while working better, and how to set it up before an exam.

What most students do

In a survey of 177 college students, Karpicke, Butler and Roediger (2009) asked people to list their study strategies. 84% listed rereading, and 55% named it as their number one strategy. Only 11% (19 of 177) said they practised recalling material, for example by testing themselves.

So if you mainly reread, you are in the majority. The question is whether the majority is right.

The core experiment

Roediger and Karpicke (2006) gave students two short prose passages, one about the Sun and one about sea otters. After a first reading, one group read the passage again, and another group wrote down everything they could remember, with no feedback.

Then came a final test, either 5 minutes, 2 days or 1 week later.

• After 5 minutes, rereading won: 81% recalled versus 75%.
• After 2 days, testing won: 68% versus 54%.
• After 1 week, testing still won: 56% versus 42%.

A second experiment pushed it further. Students either read the passage four times, read it three times and recalled it once, or read it once and recalled it three times. One week later, the four-times readers recalled 40% of the passage, while the read-once, recall-three-times group recalled 61%. The readers had gone through the text about 14 times in total; the recall group about 3 times.

One more detail matters. The students who only reread were more confident that they would remember the passage in a week. Their confidence pointed the wrong way.

Stopping too early

A common rule is "study until you can recall it, then move on." Karpicke and Roediger (2008) tested that rule with 40 Swahili-English word pairs.

All groups learned the list to the same level. Then, in some conditions, a pair was dropped from further testing once a student had recalled it. In other conditions, it stayed in the tests.

A week later, students who kept being tested on the whole list recalled about 80% of the pairs. Students whose pairs were dropped from testing recalled 36% and 33%, even though one of those groups kept restudying every pair. Repeated studying after the first correct recall added almost nothing. Repeated retrieval added a lot.

And again, the students could not tell. Every group predicted it would recall about half of the pairs.

Not only for vocabulary

A fair objection: word pairs and short passages are not a university exam that asks you to explain and apply.

Karpicke and Blunt (2011) used science texts and a final test with two kinds of questions: ones answered directly in the text, and inference questions that required connecting several ideas. They compared retrieval practice with concept mapping, a technique that is widely recommended because it makes you organise the material.

One week later, retrieval practice scored 0.67 and concept mapping 0.45, about 50% better. In a second experiment, 101 of 120 students did better after retrieval practice, and the advantage held even when the final test asked students to draw a concept map. Students predicted the opposite ranking.

The wider evidence

Single experiments can mislead, so it helps to look at reviews.

Adesope, Trevisan and Sundararajan (2017) pooled 272 effect sizes from 118 articles, with more than 15,000 participants. Practice tests beat restudying with a moderate effect (Hedges' g = 0.51). The effect appeared with and without feedback (g = 0.63 and 0.60), multiple-choice and mixed-format practice tests produced large effects (g = 0.70 and 0.80), and the benefit was larger when the practice test had the same format as the final test.

Dunlosky and colleagues (2013) reviewed ten common study techniques. Practice testing was one of only two rated "high utility" (the other was spacing, the subject of the next post). Rereading and highlighting were rated "low utility".

Why it feels worse

Testing yourself is uncomfortable. You hit blanks. You realise a topic you "knew" is half there. Rereading never shows you that, which is exactly why it feels better.

The 5-minute result above explains the trap. In the short run, rereading really does produce higher scores, so the quick check you do right after studying rewards the weaker method. The gap only turns around after a delay, and an exam is always after a delay.

How to use it before an exam

You do not need special tools. A blank page and a clock are enough.

1. Read once, then close the material. Testing works after a first exposure; you cannot retrieve what you never encoded. In the experiments above, students always read first.

2. Write a brain dump. Take a blank page and write everything you remember about one topic: definitions, formulas, the steps of a method, an example. Set a timer for 10 minutes and do not peek.

3. Check and mark the gaps. Open your notes and compare. Mark what was missing or wrong in a different colour. That list is your study plan for the next session, not the parts you already produced.

4. Turn headings into questions. "Price elasticity of demand" becomes "What is price elasticity, how do I compute it, and what makes demand elastic?" Answer out loud or on paper before checking.

5. Use past exams under exam conditions. They test retrieval in the format you will face, and the meta-analysis found larger benefits when practice and final test formats matched. Mark them honestly.

6. Keep correct items in rotation. After the Swahili result, do not drop a question the first time you get it right. Bring it back in later sessions, with longer gaps each time.

7. Check your answers. In the meta-analysis feedback did not change the size of the effect, but a self-test only shows you your gaps if you compare your answers with the source.

A concrete example

Say you have a statistics exam in ten days and the topic is hypothesis testing.

Instead of reading the lecture slides a third time, you write down from memory: what a null hypothesis is, what a p-value means, the steps of a t-test, and when you would use a paired versus an independent test. You get the steps right, give the p-value an incorrect definition, and draw a blank on the paired test.

Now you know where your next 30 minutes should go. Two days later you answer the same questions again on a blank page, plus two new ones.

Common objections

"I don't know enough yet to test myself." Then read once first. Testing comes after initial study, not instead of it.

"It takes longer." Sometimes it does: in the second experiment of Roediger and Karpicke, each recall test lasted 10 minutes and each reading period 5. But Karpicke and Blunt (2011) matched learning time exactly, and retrieval practice still came out ahead. Time spent rereading is not time saved if it does not survive the week.

"Multiple choice is not real retrieval." In the meta-analysis multiple-choice practice tests also produced large effects. Use whatever format you can make or find, and mix them.

The short version

Rereading tells you the material looks familiar. Testing tells you whether you can produce it, and the act of producing it strengthens the memory. When an exam is days away, spend most of your study time retrieving, not rereading.

If you want a tutor to do the quizzing, Learnode asks you questions in chat and keeps a record of which concepts you have recalled and when.