Learning By Teaching Improves Tutor Learning¶

Claim · All claims
Evidence · 2 studies · 2 quant-synthesis r? · q3 · i2 medium
Currency · newest study 2019 · checked against its sources 2026-10-07 (automated, 2 of 2 entries) · not reviewed by a person

Preparing to teach and teaching improve the tutor's own learning. (Whether tutors gain more than their tutees is often said but is not established by the evidence below, which does not compare the two.) Preparing to teach, explaining aloud, and responding to a learner's questions drive the tutor to organize, elaborate, and fill gaps in their own understanding.

Subclaims¶

q3 i2 Across 28 studies, teaching after preparing to teach improved the tutor's domain knowledge relative to studying without a teaching expectancy (g = 0.56), with preparing-to-teach alone giving a smaller benefit (g = 0.35) and interactive teaching outperforming non-interactive teaching. → Kobayashi 2019

q3 i? In a meta-analysis of 65 evaluations of school tutoring programmes, the children who served as tutors gained a better understanding of the tutored subject matter, alongside the gains of the tutees; the abstract reports no effect size for tutors. → Cohen et al. 1982

Evidence¶

Kobayashi 2019¶

Kobayashi, K. (2019). Learning by preparing-to-teach and teaching: A meta-analysis. Japanese Psychological Research, 61(3), 192–203. doi:10.1111/jpr.12221

q3 · meta-analysis · i2 · medium effect, g=0.56 (teaching with preparing-to-teach); g=0.35 (preparing-to-teach alone) · n=28 studies · quant-synthesis · r?

This meta-analysis pooled 28 studies comparing learners who prepared to teach, or prepared and then actually taught, against learners who simply studied the same material with no expectation of teaching. Preparing to teach produced a small-to-medium gain in domain knowledge, and preparing and then teaching produced a larger, medium gain; both helped deep as well as surface learning and persisted after a delay. The benefit was larger when the teaching was interactive (responding to a learner) than when it was non-interactive, and even the mere expectation of interactive teaching helped more than expecting to teach non-interactively. This supports the page's distinction between "prepare to teach" as a weaker intervention and "teach and respond to questions" as a stronger one. Read from the abstract only.

Cohen et al. 1982¶

Cohen, P. A., Kulik, J. A., & Kulik, C.-L. C. (1982). Educational outcomes of tutoring: A meta-analysis of findings. American Educational Research Journal, 19(2), 237–248. doi:10.3102/00028312019002237

q3 · meta-analysis of programme evaluations · i? · no effect size reported in the abstract · n=65 studies · quant-synthesis · r?

This meta-analysis synthesised 65 independent evaluations of school tutoring programmes. Besides benefiting the students who were tutored, the programmes had positive effects on the children who served as tutors: tutors came to understand the tutored subject matter better and developed more positive attitudes toward it. Tutoring had little or no effect on the self-esteem of either tutors or tutees. The abstract does not compare the size of tutor gains with tutee gains, so it does not bear on the page's statement that tutors often learn more than tutees. Read from the abstract only.

Studies bearing on this claim¶

Does preparing to teach a part, and then teaching it, improve the teaching learner's own understanding? This table gathers the studies behind this claim and the 3 narrower claims that bear on it: 5 distinct studies (2 quant-synthesis, 2 causal, 1 associational), q2–q3, 2 of 5 with an effect size coded. One row per study, codes as each entry records them; no effect is averaged and no verdict is drawn. Built by scripts/synthesize_claim.py from eval/synthesis/; a row that looks wrong is fixed on the claim it comes from.

Discussion¶

Why teaching works on the tutor. The expected effect is usually explained by the preparatory and generative demands of teaching: anticipating a learner's questions forces deeper processing, and explaining aloud exposes gaps that silent study conceals. This places learning-by-teaching close to other generative activities — see Active learning improves exam performance — in that the benefit accrues from what the tutor must retrieve, organize, and articulate, not from the social role itself. Explaining aloud is also a form of self-explanation, which links this claim to the broader family of elaboration effects; the tutor must reconstruct material in their own words rather than reread it.

Expectation matters. A plausible moderator is whether the tutor expects to teach. If the tutor does not actually teach afterward, much of the preparatory deep processing may not occur, so the effect should be strongest when teaching genuinely happens. Designers should therefore treat "prepare to teach" as a distinct, weaker intervention than "teach and respond to questions."

Boundary conditions. The effect should depend on the tutor having enough baseline knowledge to explain meaningfully; a tutor with severe knowledge gaps may simply rehearse misconceptions, and the tutee inherits those errors. The quality of the tutee's engagement (asking questions, requesting explanations) is also likely to moderate how much gap-filling the tutor performs — a passive tutee reduces teaching to recitation, stripping out the retrieval and elaboration that drive the effect. Tutee confusion or probing questions may be the single most valuable feature of the interaction for the tutor's own learning, since they force the tutor to confront and repair incomplete understanding.

Design implications. Peer-tutoring and teach-back formats should be structured so the tutor must retrieve and reconstruct rather than read aloud: assign the tutor material they have studied but not mastered, prompt the tutee to ask genuine questions, and build in a step where the tutor checks the tutee's understanding and diagnoses errors. Reciprocal role rotation (each pair member alternates tutor/tutee) lets both learners capture the tutor-side benefit while limiting the time any one learner spends explaining material they barely know. Where full peer tutoring is impractical, structured explanation tasks — teach-backs, jigsaw expert roles, or explaining to a hypothetical novice — approximate the generative demands without requiring a real tutee, though the absence of genuine probing questions likely weakens the gap-filling component. Because explaining imposes its own generative load, tutors with weak baseline knowledge may need cognitive load management supports (e.g., reference sheets, scripted prompts) so that the effort goes into reasoning rather than coping; the same expertise-sensitivity is captured by the expertise reversal effect, which predicts that supports helpful for novice tutors become redundant for knowledgeable ones.

Open questions. Two meta-analyses are recorded above, both read from abstracts: Kobayashi (2019; 28 studies, g = 0.56 for preparing and teaching, 0.35 for preparing alone) and Cohen et al. (1982; 65 tutoring-programme evaluations, tutor gains reported without an effect size). Neither compares tutor with tutee gains or teaching with equal-time alternatives. Key open

questions include whether the effect holds across age groups and domains, how it compares to equivalent-time generative activities such as writing summaries or self-explaining, and whether structured tutoring protocols (e.g., scripted roles, tutee question prompts) amplify or dampen the tutor-side benefit.

Merged from “Learning By Teaching Improves Learning” (learning-by-teaching-improves-learning): Mechanisms. The learning-by-teaching effect is typically attributed to two complementary processes. First, preparation to teach drives deeper processing: expecting to explain material to someone else prompts learners to organize content, identify gaps, and rehearse retrieval rather than passively reread. Second, actual teaching forces generative elaboration — the tutor must reconstruct explanations, answer questions, and monitor a tutee's comprehension, which exposes the tutor's own misconceptions and triggers repair. Both routes are consistent with generative-processing accounts in Cognitive Load Theory and with the constructive character of Constructivism.

Preparation vs. enactment. A recurring open question is how much of the benefit comes from expecting to teach versus teaching itself. In the one synthesis recorded here (Kobayashi 2019), preparation alone gave a smaller benefit (g = 0.35) than preparing and then teaching (g = 0.56), and the enactment adds benefits mainly when tutors receive feedback, must answer questions, or must adapt explanations to a tutee's needs. Designers should therefore not assume that merely assigning a teaching role guarantees deep processing; the task must require retrieval, organization, and explanation.

Boundary conditions. The effect is strongest when the to-be-taught material is conceptually demanding, when tutors have adequate baseline knowledge to explain it, and when teaching involves interactive dialogue rather than one-way presentation. Teaching without preparation, or teaching procedural material the tutor has not yet automatized, can produce shallow "illusion of knowing" effects or overload the tutor — a direct risk under Cognitive overload degrades learning. Structured formats — reciprocal peer tutoring, teach-backs, and jigsaw-style Collaborative Learning arrangements — tend to manage these risks better than unstructured "explain it to your partner" prompts.

Relation to other effects. Learning by teaching overlaps with Active Learning as a broad category, but the specific claim here is that the teaching role itself — not merely activity or interaction — drives the benefit. It also connects to retrieval practice and elaboration: preparing an explanation is, in effect, a demanding retrieval and organization task, linking the effect to Retrieval practice improves retention. Where the tutor is a novice explaining to a peer, the interaction also resembles the reciprocal dynamics described in Reciprocal Teaching, though that pattern emphasizes structured dialogue roles rather than the tutor's learning gain per se.

Open questions. Key unresolved issues include: (a) the relative contribution of expectancy versus enactment, (b) whether benefits accrue equally to tutees and tutors, and (c) how tutor expertise moderates the effect — very low-knowledge tutors may lack the resources to explain at all, while high-knowledge tutors may gain little from restating what they already know. Kobayashi (2019) bears on (a): preparing and then teaching (g = 0.56) gave more than preparing alone (g = 0.35), and interactive teaching more than non-interactive; neither recorded synthesis settles (b) or (c).

Merged from “Learning By Teaching Improves Mastery” (learning-by-teaching-improves-mastery): Mechanism. The expected benefit rests on two complementary processes: (a) expectation-driven preparation — anticipating that one will teach prompts deeper, more organized encoding than anticipating a test; and (b) retrieval and explanation during teaching itself, which forces the learner to reconstruct, verbalize, and repair gaps in understanding. The "protégé effect" literature suggests the expectation alone contributes meaningfully, so designers should not assume the benefit requires an actual audience. Teaching is also a maximally generative activity, consistent with the broader finding that active learning improves exam performance [+M].

Boundary conditions. The effect is most plausible when the to-be-taught material is conceptually structured (explanatory rather than rote), when learners have adequate base knowledge to explain it, and when teaching tasks require generative elaboration rather than reading a script aloud. If the teaching task can be satisfied by recitation or shallow paraphrase, the mastery benefit should shrink. Learners low in prior knowledge may also experience overload when asked to explain before they can comprehend — sequencing study before teach is safer than teach-before-study, consistent with cognitive load theory. This mirrors the expertise-reversal pattern documented for other generative methods: activities that help knowledgeable learners can burden novices.

Relation to retrieval practice. Explaining material to another person is a generative retrieval event, so part of any observed benefit may overlap with the well-established effect of retrieval practice improves retention [+S]. Well-designed comparisons match study time and retrieval demands across conditions; neither meta-analysis recorded here makes that comparison, so the unique contribution of the teaching frame beyond retrieval practice remains open.

Design implications. Where the claim is used, structure teaching tasks so they demand explanation, justification, and error correction rather than recitation; provide source material and adequate study time before the teaching episode; and consider "teach-to-prepare" manipulations (learners told they will teach, then tested without teaching) as a low-cost variant when logistics of real audiences are prohibitive. Peer tutoring formats can operationalize the claim at scale — see collaborative learning improves outcomes [+M]. Reciprocal role-reversal structures (each tutee later tutors) distribute the generative burden and prevent the tutor/tutee asymmetry from becoming a fixed ability signal within a class.

Open questions. How durable the benefit is over long retention intervals, how it compares to equivalent-time retrieval practice, and whether benefits transfer to far transfer rather than near recall all remain under-established: Kobayashi (2019) reports that benefits persisted after a delay and reached deep as well as surface learning, but neither recorded synthesis tests long intervals, far transfer or an equal-time retrieval comparison.

It is also unresolved whether the expectation-to-teach effect holds when learners repeatedly expect to teach but are never asked to — a common outcome in classroom implementations where teaching episodes get cut for time.

Merged from “Teaching Others Generative Strategy” (teaching-others-generative-strategy): Mechanism. The expected benefit comes from generative processing: anticipating explanation to another person forces learners to identify main ideas, build connections, and monitor their own understanding. This aligns with the broader family of generative strategies described in Learner highlighting of text gives a small memory benefit but no reliable comprehension benefit, and evidence on other forms of annotation is not yet recorded here and Activation improves learning. Teaching also sits within the retrieval family — explaining from memory is a form of retrieval practice, which is where much of its effect likely originates.

Expectation vs. actual teaching. A key moderator in the literature is whether learners actually teach or merely expect to teach. Expecting to teach can induce deeper study, but some findings suggest that the expectation alone, without retrieval or interaction, may encourage shallow re-reading rather than genuine organization. (Kobayashi 2019, recorded above, found preparing to teach alone still beat studying with no teaching expectancy, g = 0.35, though by less than preparing and teaching.) Actual teaching, especially with questions or confusion from the tutee, is more likely to drive the elaborative benefit. Designers should therefore build in a genuine audience — a peer, a tutee, or even a simulated one — rather than relying on the mere instruction "study this as if you will teach it."

Boundary conditions. Benefits should be strongest when the material is conceptually demanding enough to reward restructuring, when learners have adequate base knowledge to explain coherently (otherwise explanation may collapse into parroting), and when the teaching episode includes retrieval and feedback rather than reading aloud from notes. For very low-knowledge learners, the demands of composing an explanation may add extraneous load — see Cognitive overload degrades learning. This parallels the expertise-reversal pattern documented for other scaffolds: the strategy that helps a moderately knowledgeable learner restructure content can overwhelm a novice.

Comparison with related strategies. Teaching others is the interpersonal analogue of Prompting learners to self-explain improves understanding and problem solving on immediate tests by a moderate average amount, with little evidence yet on delayed or classroom gains., but adds social accountability and the possibility of tutee questions that expose gaps. Whether the social component adds benefit beyond explaining aloud to oneself is an open empirical question that this page's evidence base has not yet resolved. Peer teaching is frequently embedded in Collaborative learning improves outcomes structures, which makes isolating the specific contribution of the teaching role difficult.

Open questions. Comparisons of teaching with other generative strategies (e.g., self-explanation, summarizing) are not made by the evidence recorded above.

Related Claims¶

• Active learning improves exam performance — teaching is a paradigmatic generative activity; both claims rest on the advantage of producing over receiving.
• Case-based learning improves exam performance — like teaching, case work requires learners to organize and apply knowledge in context.
• Cognitive load management — explaining imposes generative load on the tutor; baseline knowledge determines whether that load is productive or overwhelming.
• Cognitive load reduction improves learning — scaffolds for tutors (reference sheets, scripted prompts) manage the generative load of explaining.
• Learner highlighting of text gives a small memory benefit but no reliable comprehension benefit, and evidence on other forms of annotation is not yet recorded here — like teaching, annotating is a generative activity that forces learners to transform and restate material in their own words.
• Collaborative learning improves outcomes. — peer tutoring and reciprocal teaching are structured collaborative formats in which the teaching role is assigned.
• Cognitive overload degrades learning. — unprepared tutors can be overloaded by the demands of explaining, a key constraint on the effect.
• Retrieval practice improves retention. — preparing to teach functions as a demanding retrieval and organization task, a proposed mechanism for the effect.
• Cognitive load theory — explains why teach-before-study can overload novices
• Drawing improves STEM learning by a small-to-moderate amount over not drawing and over other active study strategies, in one meta-analysis — related
• The FTS proposes that fading distinguishes expert from non-expert peers via a game-progress threshold, and peer-tutoring collaboration is expected to improve non-expert game progress and knowledge gained — related
• Tutoring benefits both tutors and tutees — related
• Peer-assisted learning interventions improve achievement for elementary students — related
• The quality of feedback students give peers predicts gains in their own work, but no recorded study isolates a causal benefit of the assessor role — related
• Structured Peer Tutoring Outperforms Unstructured — related
• Activation improves learning — teaching requires activating and organizing prior knowledge
• Prompting learners to self-explain improves understanding and problem solving on immediate tests by a moderate average amount, with little evidence yet on delayed or classroom gains. — explaining to oneself is the intrapersonal analogue of explaining to others
• A descriptive report defines peer assisted learning and asserts, without a tested result, that it promotes knowledge construction, collaboration ability and interpersonal interaction — a broader claim this one bears on
• Sociocultural learning theory holds that knowledge develops through engagement in social interaction with others, so English learners benefit from meaningful interaction with peers — related
• Non-interactive teaching is more than a special case of retrieval practice, with generative processing and social presence linked to its effects — related
• Teaching expectancy may moderate the non-interactive teaching effect, but this moderation often fails to replicate — related
• Multi-turn teaching by LLM teacher agents yields larger student gains than knowledge-point exposure alone — related
• Pedagogical experts tutoring in biology generated less deep learning than less pedagogically expert, more interactive tutors — related
• The review reports that college tutors who espoused a malleable view of intelligence to their tutees improved their own academic performance, with benefits for tutees as well — related
• Teaching (writing a verbatim lecture script) produces better create question generation than explaining (writing a textbook explanation) — related