A test that teaches
Most people think of a test as a measurement that comes after the learning. Henry Roediger and Jeffrey Karpicke open their 2006 paper in Psychological Science with a different view: "Taking a memory test not only assesses what one knows, but also enhances later retention, a phenomenon known as the testing effect." Robert Bjork's lab at UCLA makes the same point: "tests can also provide opportunities for learning."
Retrieval practice means using the effect on purpose. Anything that makes you produce an answer from memory counts, from a flashcard to a blank page on which you write what you remember. Reading the answer and nodding along does not.
The studies behind it
| Study | What people did | What they found |
|---|---|---|
| Roediger and Karpicke, 2006, Psychological Science | 180 students learned a prose passage in four short periods: four of studying, three of studying and one recall test, or one of studying and three recall tests | A week later the group that studied once and recalled three times remembered 61% of the passage, and the group that studied four times 40% |
| Karpicke and Roediger, 2008, Science | Students learned 40 Swahili-English word pairs; once a pair had been recalled, it stayed in both, or was dropped from further study, further tests or both | Pairs that stayed in the tests were recalled about 80% of the time a week later, pairs dropped from testing 36% and 33% |
| Karpicke and Blunt, 2011, Science | Students read a science text, then either drew a concept map with the text in front of them or practised recalling it, for the same amount of time | Recall beat concept mapping a week later, inference questions included; in a second experiment 101 of 120 students did better after recall |
| Adesope, Trevisan and Sundararajan, 2017, Review of Educational Research | A meta-analysis of practice tests against restudying and other ways of spending the time | Practice tests helped learning more than restudying and every other comparison condition |
In the 2006 study, the students who recalled the passage three times had read it only 3.4 times on average, against 14.2 times for those who kept rereading, and a week later they remembered about half as much again. On a test five minutes after studying, the order was reversed. Rereading wins in the short run, which is why it feels like the better method.
Why it works
Researchers still debate the exact mechanism. Several findings point the same way.
- Retrieval is its own kind of processing. Karpicke and Blunt argue that it strengthens learning by retrieval-specific mechanisms, separate from elaborative study, and may sharpen the cues that lead back to one piece of knowledge.
- Effort counts. Rowland's meta-analysis found support for effortful processing as a cause: producing an answer helped more than picking it out from a set of options.
- Even a miss helps. Nate Kornell, Matthew Hays and Robert Bjork asked people questions nobody could answer, such as "What peace treaty ended the Calumet War?", a war that never happened. Trying first and then seeing the answer led to better learning than reading the question and answer together.
- The answer has to follow. In a 2005 study by Harold Pashler and colleagues, 258 people learned Luganda-English word pairs, and showing the correct answer after a wrong response mattered far more a week later than feedback after a right one.
- Difficulty is part of it. Bjork named conditions of this kind desirable difficulties in 1994: they make practice feel harder and leave stronger long-term learning. Testing in place of restudying is one of his examples, along with spacing.
How to use it
- Close the source: put the book, video or post away before you try.
- Produce the answer: say or write it in your own words before you look at any options or hints.
- Check right away: look at the real answer as soon as you have tried. That is when a wrong answer gets corrected, as in the Pashler study.
- Grade honestly: count a half-remembered answer as half. A generous grade tells you to wait longer than you should.
- Keep testing what you know: one correct answer is where the practice starts, as the 2008 word-pair study showed.
- Space it out: spread the quizzes over days, with longer gaps as an answer gets easier.
- Ask why as well as what: Karpicke and Blunt's students gained on inference questions too, so add questions about why or how something works.
Common mistakes
The methods students reach for are the ones that feel productive. In a 2009 survey of 177 college students by Karpicke, Butler and Roediger, 84% listed rereading as a study strategy and 55% named it their first choice. Only 11% said they practised recall while studying, and 1% ranked it first. The authors put this down to illusions of competence.
The 2013 review by Dunlosky and colleagues gave practice testing its highest rating, high utility. It found that most students reread and highlight, that neither habit consistently improves performance, and that "other techniques should be used in their place (e.g., practice testing instead of rereading)."
- Trusting the feeling. Familiar material feels known. In the 2006 study the rereaders were the most confident and remembered the least a week later, and in the 2011 study students predicted that recall would work worst, when it worked best.
- Peeking. Looking at the answer before trying turns a test back into rereading. The attempt is the part that helps, even when it fails.
How scrollandlearn's quiz applies it
scrollandlearn turns posts you save from 15+ platforms into lessons, and Today's quiz turns those lessons into retrieval practice. Each question is written from one of your lessons.
- The question comes first and the answer stays hidden until you tap Show answer. With no options to pick from, it is recall, the kind Rowland found does more than recognition.
- Show answer reveals the answer straight after your attempt, the feedback the Pashler study points to.
- You grade yourself Forgot, Kind of or Knew it. The grade sets when the lesson comes back, sooner after Forgot and later after Knew it; the spaced repetition guide has the full schedule.
- Knew it does not retire a lesson. It returns after a longer gap, up to a year, which keeps the testing going after the first success.
- Quiz this topic quizzes a single topic. Each topic's knowledge score counts the lessons that have come up for a quiz, answered or not, and shows what share of them you remembered: last marked Knew it and not due again yet.
Give the model a role
Tell the model who it is and what the answer should look like before you give it the task.
Key points
- Specify the output shape, length, and audience explicitly rather than hoping.
- Provide one worked example when the format matters more than the content.
- Name the constraints, including what the answer should not do.
- Ask it to flag its own uncertainty instead of filling gaps silently.
Try this
Add a role and an output format to the prompt you use most.
Remember this
Say who it is, then what you want back.
In the quiz
What two things should a prompt say before the task itself?
Filed under AI prompts
In the example, an illustration of the lesson a YouTube video about writing prompts becomes, the question asks for two things a prompt should say before the task. If you recall both, the role and the shape of the answer you want back, that is Knew it. If you recall one, Kind of is the honest grade, and the lesson comes back sooner.
Frequently asked questions
- Is the testing effect the same as retrieval practice?
- Nearly. The testing effect is the research finding that recalling material improves later memory. Retrieval practice is the study method that uses it, by quizzing yourself on purpose.
- Does it still work if I get the answer wrong?
- Yes, as long as you see the right answer afterwards. In Kornell, Hays and Bjork's study a failed attempt followed by the answer beat reading the answer straight away.
- Are multiple-choice quizzes enough?
- They help, and recall helps more. In Rowland's 2014 meta-analysis, tests that asked people to recall the answer produced larger benefits than recognition tests, where the answer is among the options.
- How is the testing effect different from spaced repetition?
- The testing effect is about how you review: by recalling. Spaced repetition is about when you review: at gaps that grow. scrollandlearn's quiz uses both.
- Do I type my answers in scrollandlearn?
- No. You answer in your head, tap Show answer and grade yourself Forgot, Kind of or Knew it.
