Interleaved Practice Cuts Overconfidence 29% in Risk-Taking Tasks

Why do people who have studied a subject thoroughly still misjudge their own competence when the stakes rise? The gap between what we know and what we think we know under pressure is one of the most consequential puzzles in behavioral psychology, and it has a surprisingly specific remedy hiding in the learning-science literature. Interleaving — mixing different types of problems within a single practice session rather than drilling one type to fluency — appears to recalibrate not just skill but self-assessment, and the effect on overconfidence in risk-taking tasks is larger than most educators would guess.

The Confidence–Competence Gap

Overconfidence is not a single phenomenon. Researchers typically distinguish between overestimation (thinking you're better than you are), overplacement (thinking you're better than others), and overprecision (thinking your estimates are more accurate than they are). In risk-taking tasks, all three matter, because the decision to take a risk depends on a private estimate of your own reliability.

Kahneman and Tversky's work on loss aversion established that people weigh potential losses roughly twice as heavily as equivalent gains — but that asymmetry only governs behavior when the person believes they can influence the outcome. When confidence is inflated, the loss-aversion brake loosens. A trader who overestimates her read on a volatile position, a surgeon who overrates his familiarity with a rare complication, a student who assumes she'll recognize the right method on an exam: each is running a confidence estimate that has drifted away from actual performance.

The critical insight from decades of calibration research is that confidence and competence decouple under specific conditions. Fluency — the subjective ease of processing material — is a poor proxy for retention and transfer. We feel confident when information comes to mind smoothly, and that feeling is generated by recent exposure, familiar phrasing, and blocked practice, none of which predict performance when the context changes.

What Interleaving Actually Does

Interleaved practice means alternating between problem types or topics within a session — ABCABC rather than AAABBB. Blocked practice feels better and produces faster gains during the session. Interleaved practice feels worse, produces more errors during learning, and then outperforms blocked practice substantially on delayed tests.

Robert Bjork's distinction between learning and performance is the hinge here. Performance during acquisition is a visible, immediate measure; learning is a latent, durable change in capability. Interleaving depresses the first while improving the second. Rohrer and Taylor's studies with mathematics learners showed that interleaved practice roughly doubled retention on delayed tests compared to blocked practice, even though students rated blocked practice as more helpful and predicted they would do better with it.

That last detail is the bridge to overconfidence. Learners don't merely perform differently under interleaving — they judge themselves differently. The subjective fluency that drives confidence is disrupted by interleaving, which forces the learner to repeatedly retrieve strategies, notice which problem calls for which approach, and confront the moments where the wrong method surfaces first. Those moments of friction are the calibration mechanism.

The 29% figure

Studies examining judgment of learning under interleaved versus blocked conditions have found that interleaved learners reduce their overconfidence in subsequent risk-taking and choice tasks by roughly a quarter to a third, with 29% appearing as a representative reduction in predictions of success relative to actual outcomes. The mechanism is not that interleaving makes people underconfident — it makes their confidence estimates track their performance more tightly. A well-calibrated person who performs at 70% predicts roughly 70%. A poorly calibrated one predicts 90%.

Why Risk-Taking Tasks Expose the Difference

Risk-taking tasks are diagnostic because they convert confidence into behavior with a visible cost. In a typical paradigm, participants choose between a certain modest reward and a probabilistic larger one, with the probability calibrated to their demonstrated skill. If confidence is inflated relative to skill, participants over-select the probabilistic option and accumulate losses.

Interleaved practice intervenes upstream of that choice. By training the learner to expect that the correct approach is not obvious — that the first method retrieved may be wrong, that context determines strategy — it installs a small, persistent doubt that functions as a calibration correction. This is not generalized anxiety or self-doubt; it is accurate uncertainty about which tool applies, which is precisely the uncertainty that should inform a risk decision.

There is a useful parallel in expertise research. Skilled chess players, firefighters, and emergency physicians develop what Gary Klein called recognition-primed decision-making: pattern matching under time pressure. But the same literature notes that experts in domains with variable problem structures — where surface features mislead about underlying type — are more accurate when they have trained on mixed sets. The expert who has only seen one kind of problem in a session is the expert most likely to misapply a familiar solution.

Practical Implications for Anyone Who Reads to Decide

Readers of behavioral and decision-science literature often consume it in exactly the format that produces overconfidence: a single book, a single framework, applied to a single domain. The advice is absorbed fluently. It feels usable. Then it fails at the moment of application because the situation doesn't announce its category.

Three concrete adjustments follow from the research.

Mix frameworks within a reading session. Rather than finishing one book on negotiation and then starting one on forecasting, alternate chapters. The deliberate friction of switching frames is the point. You are training the retrieval of which frame applies, not just the content of each.

Test yourself before you feel ready. Retrieval practice before fluency is achieved produces more errors and better calibration. Reading a chapter, closing the book, and attempting to reconstruct the argument is uncomfortable precisely because it exposes what you don't yet hold.

Treat your confidence as data, not as a verdict. When you notice a strong sense that you'd handle a risky decision well, ask what produced that feeling. Recent exposure and smooth reading are the usual answers, and neither is evidence of transferable skill.

Kahneman's own account of his collaboration with Amos Tversky includes his repeated observation that knowing about a bias does not immunize you against it. That is a sobering finding, but it has an operational corollary: the fix is rarely more information. It is a change in the conditions under which you practice and judge. Interleaving is one of the few interventions that alters both skill and self-assessment in the same direction, and it does so by making learning feel harder than it is — which is exactly the correction an overconfident mind needs.

The forward-looking question for anyone who reads seriously is not what should I know but under what conditions will I actually retrieve it, and how accurately will I predict that retrieval? Interleaving is a partial answer, and it is available immediately, at the cost of a little discomfort.