You’re forty questions into a practice block. You miss one, open the explanation, and think right, I knew that. You flag nothing and move on. But you didn’t know it. Ninety seconds earlier, with the answer hidden, you picked something else.
Once the answer is on the screen it feels obvious, and the gap you just found closes in your head without closing in your memory. The faculty that was supposed to catch it is metacognition, your ability to judge what you know and decide what to do about it. Researchers split it into two jobs. Monitoring is your running estimate of what you know. Do I have this? Will I still have it on Friday? Control is what you do with that estimate. Study it again, move on, switch methods, stop. Bjork, Dunlosky and Kornell (2013) use this framework, adapted from Nelson and Narens (1990, as cited in Bjork et al., 2013), to describe self-regulated learning, the decisions learners make about their own studying.
Control can only be as good as the monitoring that feeds it. If your gauge reads “learned” when the truth is “familiar,” moving on is a sensible decision built on a bad reading. Earlier posts kept finding that the methods that work often feel worse than the ones that don’t. So when can your sense of knowing be trusted?
Your Sense of Knowing Is an Inference
Pintrich (2002), writing about teaching, makes the stakes clear. Knowing your own strengths and weaknesses matters, he argues, but the accuracy of that self-knowledge seems to matter most for learning. He puts it this way: “If students do not realize they do not know some aspect of factual, conceptual, or procedural knowledge, it is unlikely they will make any effort to acquire or construct new knowledge.”
So how accurate is it? When you ask yourself whether you know something, it feels like reading a thermometer. You look inward and see the answer. Bjork et al. (2013, p. 428) argue that metacognition is “more like a speedometer.” A speedometer doesn’t measure speed directly. It works it out from how fast the tires turn. Your judgments of learning, your own predictions of what you’ll remember later, are worked out in the same way, from cues. One of the strongest is fluency, the sense that something is easy to process or comes readily to mind.
Fluency is often a fair guide, but lab studies described by Bjork et al. (2013) show how it can be fooled. In one, people answered trivia questions and predicted whether they could recall each answer later. Quick answers got higher predictions, yet slower answers were the ones more likely to be recalled (Benjamin et al., 1998, as cited in Bjork et al., 2013). That later test asked people to list the answers without seeing the questions again. Bjork et al. note that speed is often a sensible cue when the final test matches the practice. In another study, larger fonts raised predictions without affecting learning, while extra study trials, which helped a lot, barely moved them (Kornell et al., 2011, as cited in Bjork et al., 2013).
People also judge as if their memory will stay as it is today, a pattern Kornell and Bjork (2009, as cited in Bjork et al., 2013) called a stability bias. Their participants underestimated how much more study would help. In a study by Koriat et al. (2004, as cited in Bjork et al., 2013), predictions were the same whether people expected the test immediately or a week later, though Bjork et al. point to another study that complicates this. Once forgetting was brought to their attention, the predictions changed. People know that forgetting happens. That knowledge just doesn’t reach the in-the-moment feeling of “I’ve got this.”
Blind Spots That Cost Study Time
The first is hindsight. Bjork et al. (2013) describe a knew-it-all-along tendency, where an answer in view seems like one you knew. The night-before scan of a chapter suffers from this. Everything looks familiar because it’s on the page.
The mirror image is foresight, a will-know-it-later tendency. During study the answer sits next to the question, so it seems obvious. On the exam the question arrives alone, and other plausible answers come to mind and compete with it. A good distractor on a board-style question is built to be exactly that competitor.
The third is about methods. In several experiments by Kornell (2009, as cited in Bjork et al., 2013), nearly all participants did better after spaced study than after massed study (the same material crammed into one block), yet most judged massing more effective. In an introductory physics course, students rated interleaved homework (problem types mixed together rather than practised in blocks) as harder and judged that they had learned less from it. On surprise tests of new, harder problems, they did better after the interleaved homework (Samani & Pan, 2021). On midterms a few days later the difference was no longer reliable, possibly masked by heavy cramming.
Given the choice, learners in lab studies tend to drop items from study once they believe they’ve learned them, even though another round would still help (Bjork et al., 2013). Add stability bias and material gets retired at the point it feels known, which may be well before it is secure. The same review adds a counterpoint. Outside the lab, students report testing themselves on items many times after they can already recall them, so how often people really stop too early is less settled.
When Kornell and Bjork (2007, as cited in Bjork et al., 2013) surveyed 472 UCLA students, about 70% said they tested themselves to find out how well they had learned. Only 18% said they did it because they learn more that way. If a quiz is only a readout, a good score means you can stop. If the quiz is also where learning happens, as the retrieval practice post argued, a good score is a reason to schedule the next one.
How to Get an Honest Reading
Picture studying a made-up pair, ocean–tree. Later someone asks how sure you are that you’ll remember it. In two word-pair studies, some people were asked right after studying and others after a delay (Nelson & Dunlosky, 1991, and Dunlosky & Nelson, 1992, both as cited in Bjork et al., 2013). Some saw both words when they judged, others only the first. Only the delayed, first-word-only condition made people actually try to pull the answer from memory, and its judgments were far better at separating pairs they would later recall from pairs they would forget. Bjork et al. note one complication they call unsettled. The test built into that judgment strengthens memory too, so part of its apparent accuracy may come from the test itself.
Their advice to a student who did worse than expected is plain: “Take a meaningful self-test without checking the answers until you are done” (Bjork et al., 2013, p. 437). Using delayed checks to decide what to study is my inference from these findings, not something the studies tested.
Decide what is “done” the next day. The end of a session, when everything feels fluent, is the worst time to judge.
Commit before you check. On flashcards or practice questions, say or write your answer, and how sure you are, before revealing anything.
Treat confident errors as the most useful items on the page. Feedback is especially effective after errors made with high confidence, a hypercorrection effect (Butterfield & Metcalfe, 2001, as cited in Bjork et al., 2013). Bjork et al. report it has been replicated many times, including with delays long enough to matter in real courses. In those studies the correct answer followed the error, and when producing errors helps rather than spreading them is still an open question.
Trust your record over your impression. Keep your percent correct per block or deck, and if it disagrees with your gut, go with the numbers.
If you teach or design courses, the gauge problem is yours too. Bjork et al. (2013) note that teachers can make the same error of ignoring forgetting when they judge what students know. They suggest that homework, weekly exams and cumulative finals can push students toward spaced study. Green et al. (2018), writing about health professions education, suggest that educators share the data on retrieval practice with students, or invite them to try it and compare their own retention against rereading. Teaching monitoring also seems to pay off. In a meta-analysis of self-regulation training described by Hattie and Donoghue (2016), teaching planning, monitoring and evaluation was linked to better achievement. So was teaching several other strategies, and the outcome was performance, not the accuracy of students’ self-judgments. Treat this secondhand report as a direction more than a measurement.
Where the Evidence Runs Out
Most of these studies use word pairs, trivia and short passages, and Bjork et al. (2013) list monitoring of complex material as an open problem. Experience is also not always a poor guide. People judged how hard anagrams were more accurately after solving one themselves than when the solution was printed beside it (Jacoby & Kelley, 1987, as cited in Bjork et al., 2013). Where there is no better basis for judgment, Bjork et al. conclude, subjective experience can be an especially good guide. And this literature is about predicting retention, remembering what you studied. It says little about whether better monitoring helps you apply knowledge to new problems.
None of these sources measures how well physicians judge their own knowledge, so what follows is my extrapolation. A presentation you have seen many times feels known, and reviewing a case after the outcome is known is the knew-it-all-along tendency in a clinical setting. Madan (2023), writing for medical educators, calls the gap between how strategies feel and how well they work a “metacognitive discrepancy.” The habit carries over to the wards. Commit to your read of the film or the lab before you see the attending’s, and keep track of the calls you were sure of and got wrong.
What This Looked Like for Me
The change that turned my first year of medical school around was less a new strategy than a new instrument. For my first few exams, every signal I had said I was prepared. I’d seen the material many times, and it looked familiar on the page. Familiarity was standing in for knowledge, and I had no way to tell the two apart.
Switching to retrieval-based study gave me a reading I couldn’t argue with. Cards I’d just studied came back blank. Topics I thought I owned fell apart when I had to produce them. It was uncomfortable, and it was the first honest data I’d had. Everything after that (spaced review, using question banks as learning events, being selective about where my time went) was a set of control decisions.
The habit I’d hand to anyone is small. When you were sure and got it wrong, stop and work out why before moving on. That moment lasts a few seconds before hindsight smooths it over.
The Core Lesson
Every strategy in this series rests on decisions about what to study, when to stop, and whether a method is working, and those decisions are only as good as your sense of what you know. That sense is built largely from fluency. Judged right after study with the answer in view, it tends toward overconfidence and does a poor job of telling what will last from what won’t, while a delayed, honest attempt at recall gives a much better reading. Pintrich (2002) puts the aim plainly: to “help students make accurate assessments of their self-knowledge, not inflate their self-esteem.” Finding a gap helps most when the gap looks like something you can close, which is the subject of the next post: growth versus fixed mindset.
This post is part of a series on the science of learning. Each post covers one evidence-based principle and how to apply it to your own studying. Follow us on Instagram @acemeded to keep up with future blog posts and related content
References
Bjork, R. A., Dunlosky, J., & Kornell, N. (2013). Self-regulated learning: Beliefs, techniques, and illusions. Annual Review of Psychology, 64, 417–444. https://doi.org/10.1146/annurev-psych-113011-143823
Green, M. L., Moeller, J. J., & Spak, J. M. (2018). Test-enhanced learning in health professions education: A systematic review: BEME Guide No. 48. Medical Teacher, 40(4), 337–350. https://doi.org/10.1080/0142159X.2018.1430354
Hattie, J. A. C., & Donoghue, G. M. (2016). Learning strategies: A synthesis and conceptual model. npj Science of Learning, 1, Article 16013. https://doi.org/10.1038/npjscilearn.2016.13
Madan, C. R. (2023). Using evidence-based learning strategies to improve medical education. Medical Science Educator, 33(3), 773–776. https://doi.org/10.1007/s40670-023-01798-9
Pintrich, P. R. (2002). The role of metacognitive knowledge in learning, teaching, and assessing. Theory Into Practice, 41(4), 219–225. https://doi.org/10.1207/s15430421tip4104_3
Samani, J., & Pan, S. C. (2021). Interleaved practice enhances memory and problem-solving ability in undergraduate physics. npj Science of Learning, 6, Article 32. https://doi.org/10.1038/s41539-021-00110-x
