Why rereading your notes feels like revision and mostly is not

The experiment that founded the field

In 2006 Henry Roediger and Jeffrey Karpicke gave 120 students a short passage of prose to read. Then half of them read it again, and half of them put it away and wrote down as much of it as they could remember, with no answers and no feedback. After that, everyone was tested: forty students five minutes later, forty two days later, forty a week later.¹

The honest version of the result starts with the part that gets left out. Rereading won. Five minutes after studying, the students who had read the passage twice recalled 81% of it, and the students who had read it once and tested themselves recalled 75%. Anyone who has ever crammed successfully for a Friday test knows this, and an article that says rereading does not work is contradicting their experience in its first line.

Then the lines cross. Two days later the rereaders were on 54% and the tested group on 68%. A week later, 42% against 56%. The group that never reread the passage still recalled, after a week, as much as the rereaders had managed after two days. Over the week the rereaders lost about half of what they had; the tested group lost about a seventh.¹

The paper’s second experiment pushed harder, with students reading a passage four times against reading it once and testing three times, and found the same shape more sharply: a week later, 61% recalled against 40%. It also asked the students to predict how well they would do. The four-times readers were the most confident group in the study, and they recalled the least.¹

What the feeling of knowing measures

That last result is the whole explanation, and it is not a character flaw. The feeling that you know something tracks how easily it comes to mind right now. Rereading is the single most efficient way to make a page come to mind right now, because the page is in front of you, and the sensation of recognising every sentence is indistinguishable from the inside from the sensation of knowing it. Recognition is cheap and leaves almost no trace. Producing the answer from nothing is expensive and leaves the trace.

So the two things come apart. The method that maximises the feeling does the least for the memory, and the method that builds the memory feels, while you are doing it, like failure. Karpicke and Blunt showed the same inversion in 2011 with a harder task and a longer horizon: students who practised recalling a science text scored about half again as much a week later as students who spent the same time building concept maps from it, on inference questions as well as recall, and they had predicted the opposite.³

The failure of judgement is not limited to choosing a method. Kornell and Bjork found that letting learners set aside the flashcards they judged they had already learned produced small but consistent decreases in later recall against keeping the whole set in rotation, and put the failure squarely in the judgement: dropping a card is only ever as good as your sense of what you know, which is the part that is unreliable.⁶

What students actually do

Three years after the crossover experiment, Karpicke, Butler and Roediger asked 177 university students how they studied.² Asked to list their strategies, 84% named rereading their notes or textbook, and 55% named it as the one they used most. Eleven per cent listed self-testing at all, and 1% named it as their main method. Given a direct choice after reading a chapter, 57% chose to read it again and 18% chose to test themselves.

The authors’ phrase for what is going on is “illusions of competence”. The students were not lazy. Most of them were choosing the method that felt like it worked, in the only way anyone can feel it, and the feeling was wrong.

The pattern is not one experiment’s. Rowland’s 2014 meta-analysis pooled 159 comparisons of testing against restudying and found testing ahead by a medium margin, g = 0.50, with the advantage larger when the practice test asked for recall rather than recognition.⁴ And the largest review of study techniques, by Dunlosky and colleagues, rated ten of them and placed rereading and highlighting among the weakest while giving its highest rating to only two: testing yourself and spreading study out.⁵

What to do instead

The replacement is not exotic and it is not more work. It is the same time, spent producing instead of recognising.

• Close the book before you decide whether you know it. Read the page once, put it face down, and write what was on it. The gap between what you thought you knew and what came out is the only honest measurement available, and it is the one rereading is designed to hide.
• Then mark it. Testing yourself without finding out whether you were right is a weaker intervention; Rowland’s pooled data found the benefit larger with feedback.⁴ Open the book after the attempt, not instead of it.
• Expect it to feel worse. A session that ends with you unsure is a session that did the work. The four-times readers in 2006 felt best and remembered least, and the feeling was the last thing that should have been trusted.
• Reread on purpose, once, at the start. The crossover says rereading is fine as a first pass through material you have never met. It says nothing kind about a fourth pass.

What this does not say

It does not say cramming is useless. The five-minute result is real, and if the test is tomorrow morning the loan does not fall due until after it. What it says is that the same method used for a subject you have to keep, across modules that build on each other, charges its interest twice.

It does not say these numbers are yours. The 2006 experiment used a prose passage recalled freely, in a laboratory, by undergraduates, and the practice test carried no feedback, which usually makes retrieval work better rather than worse. A meta-analysis of 159 comparisons is a much wider base, and the spread between studies inside it is very wide.⁴ The direction is settled. The size, for you, on your subject, is not a number anyone can give you.

And it does not say that Kerfox, or any app, has been shown to change any of this. Nobody has studied it. What can be said is that every question in the app shows the answer after the attempt and not before, and this experiment is the reason why.