Judgments of Learning: Why You Are a Poor Judge of What You Know
A graduate student I know, preparing for her biochemistry qualifying exam, spent three weeks rereading her notes until the pages felt practically memorized. On the morning of the exam she sat down confident. Two hours later she walked out pale. She could not, under pressure, reconstruct the pentose phosphate pathway from scratch, though she had read descriptions of it perhaps forty times. Her sense of knowing, it turned out, had almost nothing to do with her ability to produce the knowledge.
This gap between what students feel they know and what they can actually do is one of the oldest and most replicated findings in cognitive psychology. It also happens to be one of the most ignored, because the feeling of knowing is so immediate and convincing that it seems absurd to distrust it.
Psychologists call these subjective estimates judgments of learning, or JOLs. Janet Metcalfe at Columbia has spent a good part of her career showing that these judgments are systematically biased in predictable ways. When students study a list of word pairs and then rate how likely they are to recall each pair later, their predictions correlate only weakly with actual performance. The items that feel easy to process in the moment are not the ones that stick. And the ones that felt hard while encoding, the ones the student almost gave up on, are disproportionately remembered.
Nate Kornell and Robert Bjork at UCLA have framed this problem as an illusion of competence. Fluency, the smooth feeling of a sentence gliding past familiar eyes, is a terrible proxy for learning. Rereading produces fluency without requiring the mind to reconstruct anything. The second and third passes through a chapter feel revelatory because the material is no longer surprising, and students interpret that absence of surprise as understanding.
John Dunlosky at Kent State has documented a related phenomenon he calls stability bias. People assume that what they know now, they will know later, and what they do not know now, they will not know later. Both halves of that assumption are wrong. Forgetting is brisk and relentless, and new learning often does take hold. But the stability bias means students allocate their study time based on how things feel at the moment of judgment, not on how things will feel two weeks from now when the exam arrives.
Why does any of this matter for how a person actually studies? Because the method you choose is almost always the method your metacognition recommends, and your metacognition is lying to you.
Consider the classic lab finding from Jeffrey Karpicke and Henry Roediger, published in Science in 2008. Students learned lists of Swahili-English word pairs under different conditions. Some restudied. Some took practice tests. At the end of a training session, students who had restudied rated their expected recall much higher than the students who had been tested. A week later, the tested students recalled roughly 80 percent of the words. The restudied students recalled about 35 percent. The group that felt less confident ended up knowing far more. This is the textbook case for the retrieval practice effect, and it sits downstream of a deeper problem: students cannot feel the future, only the present.
The same pattern shows up in laboratory studies of massed versus spaced practice. Cramming feels productive. Within a single session, you can watch your performance climb. Spacing study across days feels worse, because each return to the material begins with that uncomfortable sense of partial forgetting. Students asked to choose between the two regimes reliably pick the one that feels better and produces less durable learning. This is why Bjork’s concept of desirable difficulties was so unsettling when it first circulated in the 1990s. It argued that the subjective signs of effective studying are often inverted: the feeling of ease predicts forgetting, and the feeling of struggle, handled properly, predicts retention.
One response to all of this is to try harder at introspection. This does not work. Metcalfe and her colleagues have tested a range of strategies meant to help students calibrate their JOLs, and most produce only modest improvements. You cannot out-think a bias by knowing it exists; anyone who has ever overpaid for a stock they were warned about can confirm as much.
What does work is the design of the study session itself. If your method forces you to retrieve rather than reread, your judgments of learning become more honest by accident. You notice the gaps because the gaps stop you. A blank page with a prompt and a timer cannot flatter you. This is also why drawing a diagram from memory tends to outperform rereading the diagram in a textbook: the act of production exposes what is not there.
It is worth saying that not all feelings of fluency are misleading. Sometimes you really do know the material, and your confidence is appropriate. The problem is that the cue is unreliable; it is right often enough to feel trustworthy and wrong often enough to wreck an exam. A useful rule of thumb, borrowed from Dunlosky, is to treat confidence as information about your recent experience with the material, not about your future performance.
There is also a load-management angle here. Rereading is cognitively cheap in a way that makes it seductive; the page does most of the work. Any approach that shifts the effort from recognition to generation will feel harder and tire the student faster. This is partly why sensible sequencing, in the spirit of cognitive load theory, matters: a retrieval-heavy session conducted at the end of a long day produces less than the same session conducted fresh. Judgments of learning are distorted by fatigue as much as by fluency.
The grad student I mentioned at the start did retake the exam, six months later, using almost none of the methods she had used the first time. She wrote pathways on blank paper from memory. She quizzed herself without looking. She made herself uncomfortable on purpose. She passed with a distinction. The material had not become easier. Her feeling of knowing had simply, at last, been trained to mean something.
Photo via Unsplash.