Skip to content

Peer Learning

Principle · All principles
Evidence · 44 claims (22 for, 22 mixed) · 49 studies (13 quant-synthesis, 12 causal, 7 review, 5 associational, 5 qualitative, 4 design, 3 theoretical), q1–q4 · 13 of 49 report an effect size · 36 claims rest on one study

Conditional relationship

This page owns the relationship between learners of similar standing helping one another learn: one learner explains, teaches, models or judges, and another asks, answers and responds. For learners who have already studied the material (they are not meeting it for the first time in the pair), and for goals that gain from explaining and checking ideas (concept, principle and procedure knowledge, reading comprehension, writing), a structured exchange in which each learner must explain to a peer and answer that peer's questions is expected to raise individually measured achievement (immediate-performance, course or test achievement) over studying the same material alone. The expectation runs in both directions: the learner who explains is expected to gain from preparing and explaining, and the learner who is helped from the explanation, the questions and the feedback.

Two kinds of claim in the wiki test parts of this. Meta-analyses of school peer-tutoring programmes compare tutored learners, and the learners who served as tutors, with learners who had no tutoring (Peer Tutoring Improves Achievement [+S]). A meta-analysis of 28 studies compares preparing to teach, and preparing and then teaching, with studying the same material with no expectation of teaching (Learning By Teaching Improves Tutor Learning [+S]). Several of their entries were read from abstracts only, and neither compares peer learning with an equal-time alternative such as self-explaining alone; the only comparison in the wiki of the same task done with a peer and alone is a second-hand, one-sentence report that concept mapping with a peer beat concept mapping alone (ICAP concept-mapping claim [+W]), whose design is not described. No claim in the wiki tests the doses, the role rotation or the hand-over points in the default design below; those are labelled proposals.

The relationship is conditional on four things the evidence points to: structure (defined roles, trained tutors, a script for questions and feedback; structured tutoring beat unstructured [+M]), interaction (the gain was larger when the teaching responded to a learner than when it did not), the explainer knowing enough to explain correctly (peers can pass on errors; no claim measures how often), and an individual check of what each learner can now do. It is weakened where the exchange is co-presence or answer-swapping, and it does not forecast far-transfer, which no recorded claim measures, or a large delayed-retention gain: the one randomised test of peer assessment with a delayed measure found a negligible benefit on unit quizzes that persisted at the same size to the final exam (unit quizzes [+M], final exam [+M]).

How the converted siblings sit beside this: Collaborative Learning owns joint work on a shared task of any structure; Cooperative Learning and Social Interdependence own group goals with individual accountability. This page owns the narrower one-to-one exchange of explanation and response (tutoring, teaching a peer, reciprocal roles, peer assessment as a learning activity) and does not repeat their models. Peer Discussion owns commit-then-discuss on a conceptual question, specified as a routine in the Peer Instruction pattern; the Structured Peer Review pattern owns the comment-and-revise cycle on drafts; Guided Practice and Scaffolding and Fading own how support during practice is raised and withdrawn, which applies when the helper is a peer.

Default design

The earlier page asked for tasks that require learners to explain and respond, norms for accountable interaction, monitoring of explanation quality, and structures so that some learners do not dominate while others hide. Below that guidance is made concrete. The relationship as a whole has claim support (peer tutoring against no tutoring; teaching against studying); every dose, time and threshold here is an untested proposal, and each step names its evidence status.

  1. Study first, then explain. Give the pair material each learner has already met (a reading, a worked example, a lesson segment), not material the explainer sees for the first time. Untested proposal, from the earlier page's caution that peers can amplify misconceptions; the self-explanation claim [~S] notes in its own discussion that a learner without the background to explain correctly may reinforce a misconception.
  2. Prepare to teach (5–10 minutes). Each learner writes a short plan of what they will explain: the main idea, one example, and the question they expect to be asked. Preparing to teach alone gave a smaller benefit than preparing and then teaching, in one meta-analysis (Learning By Teaching Improves Tutor Learning [+S]); the minutes are an untested proposal.
  3. Explain to a partner who must respond (5–8 minutes per turn). The explainer explains without reading from notes; the partner must restate the idea in their own words and ask at least two "why" or "how" questions; the explainer answers. Interactive teaching outperformed non-interactive teaching in the same meta-analysis; explaining is the interpersonal form of prompted self-explanation [+S] (a moderate average gain on immediate tests, mostly individual and laboratory studies). The two-question minimum is an untested proposal.
  4. Swap roles. Each learner explains a different part, or the same part to a new partner, so no one is fixed as the helper or the helped. Untested proposal: the recorded tutoring claims discuss reciprocal roles but do not test them. The reason for swapping is an argument, not a result: in the ICAP framework the dominant speaker in a pair may learn more than a partner who only agrees [~W], offered by its authors as one explanation for collaboration studies that found no advantage over working alone; no study in the wiki measures it.
  5. Give the roles a script, and train it once before relying on it (one 20–30 minute session). A role card for each side: the explainer's moves (state, give an example, check the partner's restatement) and the partner's moves (restate, ask why, say what is still unclear). Model one exchange in front of the group, with one deliberate error the partner should catch. Basis: structured peer tutoring [+M] (structured programmes produced larger achievement gains than unstructured ones; the second synthesis concerns social and behavioural outcomes); rater training was the strongest moderator of peer-assessment effects in one meta-analysis (assessor-role claim [~M]). Session length is an untested proposal.
  6. Check explanation quality while pairs talk. The teacher or facilitator listens to two or three pairs per round, notes any wrong explanation, and corrects it for everyone at the end of the round with a correct model; respond to what the learner said rather than a fixed reteach. Basis: contingent scaffolding [~M] (one-to-one tutoring studies; its classroom effect is not shown). How many pairs to sample is an untested proposal.
  7. End with an individual check. Each learner answers one new item alone, or explains the idea in writing, and rates how sure they are before seeing the answer. Learners compare their answer with the criterion and note one thing to fix. Basis: self-monitoring [~M] (a review and a theoretical synthesis; no effect size). The individual check is what separates a learner's own understanding from the pair's; Cooperative Learning owns individual accountability in groups.
  8. Plan the exchange across sessions, not only one. Untested proposal for a 4–8 week arc with two or three 20–30 minute peer sessions a week: early sessions (weeks 1–2) with full role cards, modelled exchanges and the teacher sampling pairs each round; middle sessions with the partner's prompts reduced to one card ("restate, ask why"), learners choosing which part to explain, and peer assessment against the course's own criteria, added to the explaining rather than replacing it, since peer assessment's average gain is small (g = 0.31 across 54 studies; d = .12 in one large randomised college course, Peer assessment improves performance [+M], unit-quiz claim [+M]); late sessions with learners running the exchange unprompted and the teacher sampling only. Programme duration and session timing were significant moderators in the mathematics peer-tutoring meta-analysis (Peer Tutoring Improves Achievement [+S]); session length and frequency were not; the direction of the duration effect is not recorded on the claim page.
  9. Move on when the explanation is correct and the individual check succeeds. A learner who explains correctly, answers the partner's questions and gets the individual check right on two consecutive sessions moves to harder material, or becomes the explainer for a learner meeting it for the first time; one who cannot explain after two rounds goes back to studying the material (or a worked example) with the teacher before explaining again. Thresholds are untested proposals.

When a condition is not met: if no one can listen to pairs at all, give the partner a worked answer or criterion to check the explanation against, and keep the individual check; if the explainer does not yet have the content, replace step 3 with studying a worked example together and explaining it to each other line by line.

Fitting the design to a situation

The three facts that most change the decision:

  • Whether the explainer knows enough to explain correctly. A learner meeting material for the first time can pass on errors; one who has studied it gains from explaining. If unstated, ask: have the learners studied this material, and can each answer one question on it alone?
  • Who can hear the exchange and correct it, and how often. The tutoring and scaffolding evidence leans on trained tutors, scripts and monitoring; with one teacher and thirty learners, or no teacher, the structure has to be in the role cards and the check items. If unstated, ask: during peer work, who will hear an explanation, and what happens when it is wrong?
  • What each learner must be able to do alone, and when. A pair can produce a correct answer neither learner could give alone. If unstated, ask: what will each learner do unaided after the peer work, with what aids, and how long after?
If the brief says… Then change… Basis
Learners are novice and are meeting the material for the first time Do not ask them to explain unstudied content; study a worked example together first, then explain it to each other step by step; have the teacher check one explanation per pair before swapping untested proposal; the self-explanation claim [~S] notes in its discussion that learners without background may reinforce misconceptions
Learners are advanced, or some are well ahead Let the more knowledgeable explain to those meeting the material (cross-age or cross-level), with a role card so they question rather than tell; give them harder material to prepare Older peers as tutors support gains for older and younger students [+W] (a narrative policy review, no data printed); Learning By Teaching Improves Tutor Learning [+S] for the explainer's side; first-year students in problem-based learning said selected and trained senior student tutors explained more clearly and in plainer terms than staff tutors [+W] (one focus group, perceptions only)
Learners are child (primary age) Fixed short scripts (one role card per side), sessions of 15–20 minutes, roles that swap each session, a group reward built from each child's own check Peer-assisted learning improves elementary achievement [+M] (meta-analysis read as abstract; larger effects for younger, urban, low-income and minority students, and with interdependent rewards, self-referenced evaluation and autonomy); session length is an untested proposal
Learners are adult professionals in a work setting Peers explain a procedure or case from their own work to a colleague who must ask how they would handle a variant; the colleague restates; keep the individual check as a short case decision untested proposal; the recorded claims are school and university studies
Learners are English learners, or instruction is in their second language Use structured response groups and paired tutoring on the language target, with explanation allowed in the first language where the goal is content; check oral proficiency separately, as pairing alone may not move it Peer Tutoring and Response Groups positive effects on English language development [+M] (WWC, three RCTs); limit: cross-age tutoring showed no significant effect on oral proficiency [~M] (42 grade 3–5 students, daily 20-minute sessions over three months)
Goal is reading comprehension Reciprocal roles on a text: each learner in turn leads predicting, questioning, clarifying and summarising, after the teacher has modelled all four Reciprocal Teaching Improves Reading Comprehension [+M] (16 studies; effect much larger on experimenter-made tests than standardised ones)
Goal is a written draft or a product judged against criteria Peers assess each other's drafts against the criteria and each author revises; train raters on the criteria with one anchor example first; expect a small gain, so keep an individual check and do not let peer assessment replace the explaining steps; see the Structured Peer Review pattern Peer assessment improves performance [+M] (54 control-group studies, small effect against no assessment); assessor-role claim [~M] (rater training the strongest moderator; the assessor benefit itself is correlational); peer assessment in a large college statistics course [+M] (matched randomised crossover, about 300 students, d = .115 on unit quizzes against instructor grading, d = .122 on the final exam [+M]); for two or three reviews per draft, several peer reviews improved drafts more than one expert's [+W] in one 28-student experiment read second-hand, as the pattern sets out
AI feedback is available, or the brief wants peer review combined with an AI tool (writing especially) Keep the peer review and add AI as a second source after it, not instead of it: each author reads both, notes where they agree and differ, and decides what to revise; keep a teacher check on content and argument, since AI-assisted peer comments drifted toward technical detail; train peers on the criteria as above AI-peer integrated feedback beat conventional peer feedback on all four writing criteria [+M] (one quasi-experiment in L2 writing classes, d = 1.41 to 1.16 by criterion, smallest gain on coherence and cohesion; read as one study); AI-assisted peer feedback gave fewer content and argument comments than teacher feedback [~M] (one quasi-experiment, two classes); peer groups and ChatGPT-4o did not differ in feedback quality at the first stage [~M] (13 groups; the AI was rated higher at the second stage; no equivalence test); students read peers as contextual and AI as general but actionable (complementary-feedback claim [~W], perceptions claim [~W]; reflections, no learning measure)
The brief names its own routine, rubric, categories or criteria (its own check-in, prompt or template) Keep it as given and run the peer exchange on it; change only one thing the evidence says matters: train peers on that rubric once, or add the partner's "restate and ask why" move, or add one individual check untested proposal; rater training from the assessor-role claim [~M]
classroom with one teacher and a large group, or scarce facilitator time (volunteers, a busy shift, one tutor checking in briefly) Put the structure in the role cards and the check items, not in the facilitator: the facilitator samples two or three pairs per round, corrects the commonest error for everyone, and reads only the individual checks Structured Peer Tutoring Outperforms Unstructured [+M]; sampling is an untested proposal, and contingent help is shown one-to-one only (contingent scaffolding [~M]); in an IES study of mathematics curricula, teachers directing students to help one another, and asking the class whether it agreed with a student's answer, went with higher achievement [+W] (correlational, no effect size)
Mixed modes: some learners remote, some in the room Pair remote with remote and room with room, or give mixed pairs a shared document and an audio channel; schedule the peer exchange explicitly and time it, since it is the activity that tends to be dropped online In case-study courses that shifted to remote instruction, few instructors offered collaborative learning activities, and some dropped ones used before [~W] (case studies of 29 courses; describes what instructors dropped, not learning); the pairing rule is an untested proposal
Online self-paced or phone-only, no live session Asynchronous exchange: each learner records a 1–2 minute voice note or short text explaining one idea; an assigned partner must reply within a set window with a restatement and one question; the explainer answers untested proposal; the remote-shift claim [~W] shows peer activity disappears unless designed in
Learners with low confidence, low literacy, or anxious about speaking in front of others Pair with a trusted partner; let them rehearse the explanation privately first, explain orally or by showing rather than writing, use sentence starters, and keep the first rounds unassessed; the facilitator models that a wrong explanation is expected and fixable Social engagement in gateway math builds confidence, dependent on faculty fostering a supportive environment [~W] (interviews and surveys, 76 faculty and 50 students); in interviews with six Thai undergraduates after a genre-based writing course, some said they had felt uncomfortable working in pairs or groups [~W] even while crediting collaboration for their improvement; the specific moves are untested proposals
Material is private or sensitive (personal histories, health, work incidents) The learner controls what is shared and with whom; peers explain the concept on a neutral or invented example, and the learner applies it to their own case privately untested proposal
Time is one session only, or a fixed short budget (for example 30 minutes) One cycle: 3 minutes review, 5 minutes prepare, two 6-minute explain-and-question turns with roles swapped, 5 minutes individual check, 5 minutes teacher correction of the commonest error; report only immediate performance untested proposal; the tested contrast (teaching against studying) does not set a duration
A whole-task performance the brief wants kept whole (a presentation, a clinical encounter, a lesson taught) Peers watch the whole performance and debrief against the brief's criteria (one strength, one change, one question), rather than splitting it into parts to explain; where it can be repeated, run peer review on an early performance and again on a later one the debrief format and the repeat are untested proposals; for presentations, a review of college peer assessment reports peer review of oral presentation skills improving performance, and peer (not instructor) assessment of four speeches raising self-concept on the last against a control group [+W] (two older studies read second-hand, no effect sizes; the review says such studies are too few to say which feedback works best)
Learners produce for a real audience (a performance, a product for clients) Use peers as the rehearsal audience: each learner explains or performs to a peer who plays the audience and asks what that audience would ask, before the real event untested proposal
Stakes are high, or grades depend on the work Keep peer judgements formative; grade only the individual check or the individual product; record which explanations were corrected; if peer marks are used at all, spot-check a sample against the teacher's (reactive control) and do not use learners' marks of their own work untested proposal; individual accountability is modelled in Cooperative Learning; in one online exam, peers' marks averaged 0.12 points below the teacher's while authors marked their own papers 0.83 above [~W] (descriptive, no test); spot-checks and rewarding review quality are the two kinds of quality control [+W] in one inventory of practice, untested as a comparison; the Structured Peer Review pattern carries them for drafts
A misconception is widespread in the group Do not start with pair explanations; have every learner answer a diagnostic item first, address the misconception with the whole group, then use pairs to explain the corrected idea untested proposal; contingent support evidence is one-to-one (contingent scaffolding [~M])

Fifteen of the nineteen rows cite a claim, but only the advanced-learner, age, language, reading, writing, AI-feedback, large-group and whole-performance rows rest on a claim that tests a peer arrangement against an alternative (the AI-feedback row on single quasi-experiments in university writing classes, the whole-performance row on two older presentation studies read second-hand); the rest carry a neighbouring finding (self-explanation, contingent tutoring, descriptive case studies of remote teaching, interviews, descriptive marking data) or are untested proposals, and say so. Re-read 2026-10-19 (wave 2): the advanced-learner, writing, large-group, low-confidence, whole-performance and high-stakes rows gained bases from claims linked since the first re-read; none of them changed a row's move except the high-stakes row (spot-check peer marks; no self-marks) and the whole-performance row (repeat review across performances).

Observation, state and explanation

What can be observed: what each learner said when explaining (record a sentence or two), which questions the partner asked, whether the partner's restatement matched, what the facilitator corrected, and each learner's individual check. "Understands it", "was carried by the partner" and "learned a wrong explanation from a peer" are inferred states. A correct pair answer shows what the pair can do with each other's help, not what either can do alone. Nor is a learner's report that peer work helped a measure of what it did: in the randomised college statistics course, students' perceived benefit from peer assessment was uncorrelated with their estimated actual benefit [~M] (r = .01), so judge the step by the individual check, not by a satisfaction survey.

Response Plausible explanations Observation that could change the interpretation
Pair answer correct, one learner's individual check wrong The partner carried the reasoning; the learner understood in talk but cannot produce it alone; the check item differs more than intended Ask that learner to explain the idea to the facilitator; give a matched item with the partner's explanation available, then without it.
Both learners give the same wrong explanation The explainer passed on an error; both share a misconception from instruction; the material was not studied before the exchange; the less sure learner deferred to the more confident one Check each learner's answer, and how sure each is, before the exchange next time; give an item where the misconception and the correct idea predict different answers. In one experiment, the more confident a learning partner's answers, the more competent learners judged that partner [~W], so a confidently wrong explainer is likely to be believed.
One learner does all the explaining, the other only agrees Uneven confidence or status; the partner has nothing to ask; the role card is not being used Swap roles with a fresh part; require the partner's restatement before the explainer continues; ask the quieter learner privately what was unclear. Agreeing without contributing is the pattern ICAP argues leaves the listener learning less than the speaker [~W] (a theoretical interpretation, untested here), so compare the two learners' individual checks.
Explanations are fluent but repeat the text Recitation rather than reorganising; the material is too easy for this learner; the partner asks no real questions Ask a "what if" question the text does not answer; give the learner harder material to explain.
Peer comments on drafts are generic, or only about surface errors The reviewer was not trained on the criteria; the reviewer lacks the content to judge argument; an AI tool in the loop is steering comments toward technical detail Have the reviewer apply the criteria to one anchor example and compare with the teacher's reading; count content versus surface comments with and without the tool (in one quasi-experiment AI-assisted peer comments carried fewer content suggestions than teacher feedback, feedback-focus claim [~M]).
Peer work goes well in the room but stops online No scheduled time; no shared surface; the activity was replaced by an individual quiz Record whether the exchange happened at all before judging its effect; restore a timed, scripted exchange and compare.

These rows are proposals, untested with learners as diagnostics. Listening in changes the exchange it observes, so record when the facilitator joined a pair, and keep explanations open rather than choosing the one whose label fits.

Evidence and qualifications

The claims below are the model's core. They test peer arrangements against no peer arrangement or against studying alone, mostly in school and university settings, mostly at the end of a programme or session, and several were read from abstracts only.

  • Peer Tutoring Improves Achievement [+S]: two meta-analyses of school peer-tutoring programmes. In 65 evaluations, tutored learners outperformed controls on examinations and the learners who served as tutors showed the same pattern (abstract only, no pooled effect size); in 50 mathematics studies the pooled effect was small (g = 0.333), moderated by educational stage, study design, programme duration, tutor knowledge level, time of day and sample size, but not by age, assigned roles, tutee ability, session length or frequency. Its ## Studies bearing on this claim table counts 4 distinct studies, all quantitative syntheses. The claim page says neither entry shows tutors gain more than tutees.
  • Learning By Teaching Improves Tutor Learning [+S]: across 28 studies, preparing and then teaching improved the teacher's own domain knowledge over studying without expecting to teach (g = 0.56), preparing alone less (g = 0.35), with interactive teaching better than non-interactive and the benefit reported for deep as well as surface learning and after a delay (abstract only). It also cites the 65-evaluation tutoring meta-analysis for tutor gains. It does not compare teaching with an equal-time alternative such as self-explaining or retrieval practice.
  • Structured Peer Tutoring Outperforms Unstructured [+M]: in the 65-evaluation tutoring meta-analysis (read in its conference version), structured programmes produced larger achievement effects than unstructured ones, though unstructured tutoring still helped; in 36 elementary peer-assisted-learning studies, structured roles were related to the size of social, self-concept and behavioural effects. Marked M here, below its cap, because only one synthesis bears on achievement.
  • Peer assessment improves academic performance by a small average amount [+M]: 54 control-group studies, primary to tertiary, g = 0.31 against no assessment and 0.28 against teacher assessment, not different from self-assessment. It does not separate the assessor's gain from the assessee's.
  • Prompting learners to self-explain improves understanding and problem solving on immediate tests by a moderate average amount [+S]: explaining to oneself, g = .55 across 69 effect sizes, with classroom and delayed evidence much more limited. It is the mechanism this page's explaining step relies on, carried from individual prompting to explaining to a peer, and it is also the comparison the peer claims do not make: whether a partner adds anything over explaining alone is not shown in the wiki.

The claims that limit or qualify the model:

Learners, outcomes and comparisons differ across these claims (elementary peer tutoring, undergraduate teaching-expectancy studies, peer assessment across levels, university writing with AI tools), so do not rank them by effect label. None compares peer explanation with an equal-time individual generative activity in a design the wiki has read.

The evidence by question

Further evidence, not yet read against this model

Claims that bear on peer learning but do not change the model above, each with what it is and why it stays here. Read against the model on 2026-10-19: the concept-mapping, peer-assessment (Sun et al. 2015), AI-peer feedback, feedback-focus, peer-against-ChatGPT and two student-perception claims that were listed here now sit in the model (the AI-feedback situation row, the qualifications and the evidence by question). Re-read again on 2026-10-19 (wave 2): of eleven claims linked since, ten now sit in the model (default step 4, six situation rows, the observation table and the evidence by question) and one is below.

Objective and learner-valued goal

The designer's objective is usually what each learner can do alone afterwards: answer, explain or perform on a representative task, at a stated horizon. Elicit separately what learners want from working with a peer: to be told the answer quickly, to avoid looking slow in front of a classmate, to help a friend, to practise saying things aloud before a real audience. Agreement and divergence are both observations. A learner who values getting the answer may push the partner to tell rather than explain; one who values not looking slow may agree with an explanation they did not follow.

For example, a nursing student may want to be able to explain a medication change to a patient's family, while the course objective is a correct written dosage calculation. Peer teaching of the calculation serves the objective; a rehearsal in which a peer plays the family member and asks what a family would ask serves the learner's goal. Record both, so that a correct calculation is not read as the capability the learner came for.

What would revise this model?

The expectation should weaken if comparisons of peer explanation with an equal-time individual alternative (self-explaining, retrieval practice, a worked example) find no advantage for the peer arrangement; then the page should recommend explaining, with or without a partner, rather than peer learning as such. If a study measuring both partners finds the learner who only listens and agrees gains as much as the one who explains, the role swap (step 4) is a fairness rule rather than a learning one. If the explainer's gain disappears when the partner asks no questions, the interactive element is the active ingredient and the role card should centre on questioning. If studies of untrained, unscripted peer pairs show gains as large as structured programmes, the training and script steps are a preference, not a condition. If delayed tests show the peer-tutoring gains fade, the model should say "immediate" throughout. If further randomised tests of peer assessment keep finding effects near d = .1, the peer-assessment step should be framed as practice in judging quality rather than as an achievement intervention; if adding AI feedback to peer review keeps outperforming peer review alone, the AI-feedback row should move into the default design for writing. Do not protect the model by calling every failure "too little structure" after the fact.

Pair performance, each learner's immediate unaided performance, delayed retention, transfer and the learner's own valued use are separate claims. The present evidence establishes no session length, no number of rounds and no rule for when to swap roles.

Claims

  • Peer Discussion — one of the clearest vehicles for peer learning
  • Community of Inquiry — explains how peer interaction can become substantive inquiry rather than social activity alone
  • Collaborative Learning — peer learning is a narrower principle focused on learner-to-learner explanation and response
  • Cooperative Learning — group goals with individual accountability; this page's individual check borrows its accountability model
  • Social Interdependence — why helping and holding peers to account follows from a shared goal
  • Guided Practice — support during practice, here given by a peer
  • Scaffolding and Fading — how a peer helper's support should be raised and withdrawn
  • Formative Assessment — peer feedback is often a low-stakes formative mechanism before final evaluation.
  • Process-Based Writing — drafting and revision make peer review especially useful.
  • Building Empathy — constructive peer review depends on understanding audience, tone, and the impact of feedback.

Examples

  • Structured draft review: Learners use a rubric and sentence stems to comment on a peer's draft before revision.
  • Gallery walk critique: Learners review several products and leave targeted feedback linked to criteria.
  • Real-time digital peer response: Learners respond to classmates' work in a shared platform while the task is still revisable.
  • Two-stage review: Learners first identify strengths and evidence, then suggest one or two specific next moves.

Illustrative

Peer Instruction — Learners answer a conceptual question individually, discuss reasoning with peers, then revise their answer.

Jigsaw Method — Each learner becomes a temporary expert and teaches part of the whole to peers.

Peer revision or peer explanation routines — Learners respond to drafts, interpretations, or solutions and improve their own understanding through explaining what works and why.

Peer Teaching — Learners teach concepts, processes or interpretations to one another.

Structured Peer Review — Learners comment on drafts against criteria and each author revises.

Learning by Teaching — Learners prepare and deliver instruction to peers or tutees.

Key Sources

  • Topping, K. J. (2005). Trends in peer learning. Educational Psychology, 25(6), 631-645. https://doi.org/10.1080/01443410500345172
  • Fiorella, L., & Mayer, R. E. (2013). The relative benefits of learning by teaching and teaching expectancy. Contemporary Educational Psychology, 38(4), 281-288. https://doi.org/10.1016/j.cedpsych.2013.06.001
  • Leijen, D. A. J. (2017). A novel approach to examine the impact of web-based peer review on the revisions of L2 writers. Computers and Composition, 43, 35-54. https://doi.org/10.1016/j.compcom.2016.11.005
  • Lin, G. Y. (2018). Anonymous versus identified peer assessment via a Facebook-based learning application: Effects on quality of peer feedback, perceived learning, perceived fairness, and attitude toward the system. Computers & Education, 116, 81-92. https://doi.org/10.1016/j.compedu.2017.08.010
  • Wu, Y., & Schunn, C. D. (2020). When peers agree, do students listen? The central role of feedback quality and feedback frequency in determining uptake of feedback. Contemporary Educational Psychology, 62, 101897. https://doi.org/10.1016/j.cedpsych.2020.101897
  • Sun, D. L., Harris, N., Walther, G., & Baiocchi, M. (2015). Peer assessment enhances student learning: The results of a matched randomized crossover experiment in a college statistics class. PLoS ONE, 10(12), e0143177. https://doi.org/10.1371/journal.pone.0143177
  • Liu, W. (2026). AI-peer integrated feedback in second language writing classes: Exploring students' engagement and writing performance. Frontiers in Psychology, 17, 1871951. https://doi.org/10.3389/fpsyg.2026.1871951
  • Tang, J., Li, P., & Luo, R. (2026). Teacher feedback vs AI-assisted peer feedback in L2 writing: A quasi-experimental study in a Chinese university. PLoS ONE, 21(6), e0345976. https://doi.org/10.1371/journal.pone.0345976
  • Li, X., Zhang, Y., & Liu, T. (2026). Comparative analysis of peer group and AI-generated feedback in peer assessment: Insights into feedback quality and student perceptions in higher education. Australasian Journal of Educational Technology, 42(4), 129-146. https://doi.org/10.14742/ajet.11467
Page metadata
Field Value
type principle
id peer-learning
aliases peer-feedbackpeer-review, peer-feedback
title Peer Learning
description For learners who have already studied the material, a structured exchange in which each must explain to a peer and answer that peer's questions (tutoring, teaching a peer, reciprocal roles, peer assessment) is expected to raise each learner's individually measured achievement over studying alone, for explainer and helped alike; meta-analyses test peer tutoring against no tutoring and teaching against studying, while no claim tests the doses, role rotation or hand-over points.
status review
canonical True
generated by claude/unspecified on 2026-10-19
trust unverified — no one has confirmed this page's content
sources 9 — topping-2005, fiorella-2013, leijen-2017, lin-2018, wu-2020, sun-2015, liu-2026, tang-2026, li-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.