Read a worked solution and every step makes sense. Close the book, try the next problem, and the sense is gone. Self-explanation is aimed at that gap: stopping at each step to explain to yourself why it is there.
Like elaboration and elaborative interrogation, the two strategies before it, self-explanation asks you to produce something the material did not hand you. Of the three, it has the most direct evidence for transfer, meaning that learning carries over to new problems instead of only being remembered. It also takes longer, and when one study gave a comparison group about the same time for extra practice, the advantage became small and uncertain.
What Counts as a Self-Explanation
Michelene Chi, whose group ran the classic studies described below, defines self-explaining as explaining to yourself in an attempt to make sense of new information (Chi, 2000). In the way her group scored explanations, the useful ones contain an inference: something that adds information beyond what the text gives.
Her own example comes from a text on the circulatory system. A student reads that the septum divides the heart lengthwise into two sides and adds that it must keep the blood on each side from mixing. The text never said that; she inferred it.
A paraphrase usually adds nothing new. Restating “shortness of breath” as “dyspnea” is a paraphrase. It swaps one label for another without adding a mechanism or a link. That example comes from a recent medical-education study that built a tool for sorting students’ self-explanations by type (Vachon Lachiver et al., 2025).
The Eighth Graders Who Explained Every Sentence
In one classic study (Chi et al., 1989, as cited in Chi, 2000), college students who had not taken college physics studied worked examples from a mechanics textbook and were encouraged, though not required, to explain them. In another (Chi et al., 1994, as cited in Chi, 2000), eighth graders read a 101-sentence text on the circulatory system. One group was prompted to explain after every sentence; a control group read the text twice. Both studies tested learning a week later.
In both, students who generated many self-explanations learned considerably more than those who generated few. The prompted eighth graders also gained more than the read-twice group, most clearly on the hardest questions. By the end, eight of nine prompted students held the most accurate model of the circulatory system, against two of nine unprompted students.
Learners chose how much to explain, so the comparison of high and low explainers is correlational. Chi found no significant differences in prior knowledge or ability between them, in small samples, while Dunlosky et al. (2013) caution that in some studies the better explainers may also have known more. In the circulatory-system study, the one-week test was also open-book.
The most surprising finding concerns errors. About a quarter of the students’ explanations were wrong, yet the results looked the same whether or not those were counted. Most wrong explanations were later contradicted by the text, and Chi suggests this gave students a conflict to resolve. She proposed that self-explaining both fills gaps the text leaves open and repairs a learner’s working picture of the system when something does not fit.
Repair is her interpretation, built largely from one student’s explanations, and she sets its limits herself. If the starting picture is the wrong kind of picture, as when natural selection is imagined as a single causal event, patching it step by step cannot fix it. The learner needs a new framework.
Wrong explanations were harmless with a text that went on to correct them. Without such a source, an error has nothing to bump into. Noticing a misfit is also harder than it sounds. In the physics study, poor learners acknowledged failing to understand far less often than good learners did, and mostly believed they understood when they did not (Chi, 2000). That gap returns in the post on metacognition.
Is It Worth the Time?
Even the classic biology study is disputed on cost. Chi describes reading twice as taking about as long as reading once with explanations, while Dunlosky et al. (2013) report that explaining after every sentence roughly doubled study time. They also found that few studies equated time on task; in those that reported it, self-paced explaining usually took noticeably longer. Where time was equated, the results were, in their words, “somewhat mixed.”
McEldoon et al. (2013) built a randomized experiment around exactly this question. Sixty-nine children aged 7 to 10, struggling with equations such as 3 + 4 + 8 = _ + 8, were taught a procedure and assigned to one of three conditions. A control group solved six problems. A self-explanation group solved the same six and explained how and why sample answers were right or wrong. An additional-practice group solved twelve, which took a little less time than explaining did. Everyone was tested immediately and two weeks later.
Against the control group, self-explanation improved conceptual knowledge, here the idea that both sides of an equation must be equal. Its effect on procedural knowledge, the ability to carry out the steps, was less clear. Against extra practice, the differences were small enough that chance could explain them. They leaned toward self-explanation on equation structure and on transfer problems that required adapting the procedure. On problems like those practiced, the practice group scored slightly higher, again within chance. The pattern held at two weeks.
Asked why an answer was right or wrong, most children gave procedural reasons and rarely mentioned equivalence. The authors suggest that support for explaining may matter for learners this young. One trial with children aged 7 to 10 cannot settle the question for medical students, but it is the cleanest test in this post of explaining against a similar amount of extra practice.
What a Resident’s Worked Case Adds
In an earlier study (Chamberland et al., 2011, as cited in Chamberland et al., 2015), students on clinical rotations who self-explained while working through written clinical cases diagnosed new cases in the same clinical area better a week later than students who did not. The benefit appeared for unfamiliar cases only.
Chamberland et al. (2015) then asked what modeled examples and prompts add. Third-year medical students all diagnosed four jaundice cases while self-explaining aloud. One group then heard a resident’s recorded self-explanation of each case and answered prompts aloud. A second heard the recordings with silent gaps instead. A control group did word puzzles. A week later, everyone solved similar cases and cases with different, harder diagnoses within jaundice.
On the harder cases, only the group that heard examples with prompts reliably beat the control group. The group that heard the same examples without prompts did not. Justification prompts asked which pathophysiological mechanisms could explain an interpretation. Revision prompts asked, “How does this reasoning relate to yours?” The models were residents, two years ahead of the students. The authors chose them because an earlier unpublished study found little benefit from peer or expert examples; this study did not compare the three.
Because every group self-explained, the study tests what examples and prompts add on top of the habit. The control group, which only self-explained the four cases before its word puzzles, showed no gain at the end of the session yet scored higher on similar cases a week later. The design cannot say whether explaining, meeting the cases again or the week produced that, so it is a reason to hold off judging the habit by how a session ends.
What We Still Don’t Know
Whether self-explanation depends on prior knowledge is unsettled. Dunlosky et al. (2013) state that “no research has systematically examined self-explanation effects as a function of knowledge level,” and the few relevant studies point different ways. Fonseca and Chi (2011) propose that it helps novices and experienced learners differently, filling gaps for the first and repairing flawed models for the second. That mixed picture sets it apart from elaborative interrogation, whose dependence on prior knowledge is much clearer.
Transfer and retention need separate answers. On transfer, among studies using problem-solving tasks, Dunlosky et al. found effects on new problems of the same structure in “virtually every study,” and some effects on problems of a different structure, in math and pattern learning. On retention, meaning keeping what you learned over time, the record is thin: almost every study tested within minutes, and Dunlosky et al. call durability “woefully underexplored.” Like elaboration, self-explanation is about how you take material in. Keeping it over weeks and months is the job of retrieval practice and spacing.
How to Use It Tomorrow
Use it on worked examples, multi-step problems and clinical cases, where the evidence lives. At each step, ask why the step is there, and say the answer before reading the explanation or checking the key. When ready-made explanations could be looked up, the benefit shrank, presumably because learners tried little before looking (Schworm & Renkl, 2006, as cited in Dunlosky et al., 2013).
Explain as you go, during the problem set. In a logical-reasoning study, students who explained while solving each problem did better on later abstract transfer problems than students who explained after the set (Berry, 1983, as cited in Dunlosky et al., 2013). If you write your explanations, watch how many you produce: students who typed their explanations produced far fewer of them and lost the benefit (Hausmann & Chi, 2002, as cited in Fonseca & Chi, 2011).
Check every explanation against the worked solution, the answer explanation or the textbook. Two questions from the Chamberland study make a ready template: what mechanism could explain this finding, and how does my reasoning compare with the worked solution? Budget the extra time, and spend it where you need to understand or adapt a method. If your goal is speed on problems like the ones you have practiced, extra practice did about as well in the one trial that compared them.
If you teach, the Chamberland study suggests a format. Present a worked case explained aloud by someone a little further along, such as a resident for third-year students. Then ask two kinds of question: what mechanism explains this step, and how does your reasoning compare with the example?
What This Looked Like for Me
In medical school, self-explanation paid off most in pathophysiology, and the clearest case was working through clinical problems aloud. Early on I treated practice questions as a recall exercise: read the vignette, pick the answer, check it, move on. When I got one wrong, I noted the right answer and kept going, which felt efficient.
What changed was that I started explaining each step before looking at the answer. Why would this lab value be elevated, given this mechanism? If the diagnosis is what I think it is, why does this finding follow, and why would the alternative not produce it? In pharmacology, instead of memorizing that a drug lowers blood pressure, I talked through the receptor, the downstream pathway, and why the side effects I was being asked about came from that same mechanism. My explanations were often clumsy and sometimes wrong. Trying to produce them surfaced gaps I did not know I had.
I cannot run a controlled experiment on myself. I changed several strategies at once, and self-explanation came bundled with spacing and retrieval practice, so I cannot separate its contribution. What I can say is that explaining before checking changed what I noticed. After that, a wrong answer showed me where my understanding broke.
The Core Lesson
Self-explanation means adding an inference the material did not give you, and the evidence for it comes from worked examples, problems and cases. Of the three strategies so far that ask you to produce your own explanations, it has the most direct evidence for transfer, mostly to problems of the same structure, and it usually costs extra time. Against a similar amount of time spent on extra practice, the advantage can be small. Wrong explanations did little harm when a reliable text corrected them, so explain before you look, then check.
The next post turns to the Feynman Technique, which moves from explaining to yourself to explaining to an imagined listener.
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
Chamberland, M., Mamede, S., St-Onge, C., Setrakian, J., Bergeron, L., & Schmidt, H. (2015). Self-explanation in learning clinical reasoning: The added value of examples and prompts. Medical Education, 49(2), 193–202. https://doi.org/10.1111/medu.12623
Chi, M. T. H. (2000). Self-explaining expository texts: The dual processes of generating inferences and repairing mental models. In R. Glaser (Ed.), Advances in instructional psychology (Vol. 5, pp. 161–238). Lawrence Erlbaum Associates.
Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students’ learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public Interest, 14(1), 4–58. https://doi.org/10.1177/1529100612453266
Fonseca, B. A., & Chi, M. T. H. (2011). Instruction based on self-explanation. In R. E. Mayer & P. A. Alexander (Eds.), Handbook of research on learning and instruction (pp. 310–335). Routledge. https://doi.org/10.4324/9780203839089.ch15
McEldoon, K. L., Durkin, K. L., & Rittle-Johnson, B. (2013). Is self-explanation worth the time? A comparison to additional practice. British Journal of Educational Psychology, 83(4), 615–632. https://doi.org/10.1111/j.2044-8279.2012.02083.x
Vachon Lachiver, E., Chamberland, M., Bergeron, L., Setrakian, J., Chebbihi, H., & St-Onge, C. (2025). Documenting medical students’ use of self-explanations: Tool development and initial validity evidence. Canadian Medical Education Journal, 16(4), 14–22. https://doi.org/10.36834/cmej.78199
