Training · Cross-cuttingMoveprovenverified

Peer Instruction

also known as think-pair-share, ConcepTest method, Mazur method, interactive engagement

Tags: pedagogypeer-instructionthink-pair-shareConcepTestMazuractive-learning

The instructor pauses the lecture to pose a short conceptual question — a ConcepTest — that requires genuine understanding rather than recall. Learners think individually and commit to an answer (often by show of hands or clicker), then turn to a neighbor and argue for their answer, trying to convince each other. The instructor then polls again and explains. Invented by Eric Mazur at Harvard in 1991 after discovering that his physics students could pass standard tests while holding fundamental misconceptions about Newtonian mechanics. The key insight is that a student who understood a concept 15 minutes ago explains it more effectively to someone who does not yet understand it than a professor who understood it 30 years ago. The recent understander still remembers the specific confusion they had to resolve.

How the learner advances

flowchart TD A[Instructor poses ConcepTest diagnostic conceptual question] --> B[Individual think time every learner commits to answer] B --> C[First vote reveal distribution] C --> D{30–70% correct?} D -- Below 30% --> E[Direct instruction first then retry] D -- Above 70% --> F[Move on concept understood] D -- Yes: 30-70% --> G[Peer discussion turn to neighbor, argue position] G --> H[Second vote reveal new distribution] H --> I[Instructor explains addresses specific wrong intuitions] I --> J[More capable learner misconception resolved through argument]

Intent. Replace passive absorption of lectures with active sense-making by requiring learners to commit to an answer, argue for it with a peer, and update their understanding before moving on.

When to apply. Apply in any lecture or large-group teaching context where the learner population is heterogeneous — some ahead, some behind — and where conceptual misconceptions (not just knowledge gaps) are the primary barrier to understanding. Particularly effective in STEM courses where learners can do calculations correctly while holding wrong physical intuitions. Also applicable to AI training pipelines: an agent prompted to explain a concept to a simulated peer, then receive a counter-explanation, performs better than one that only receives correct answers.

Threshold — earns the next step. Second-vote correct-answer rates are consistently higher than first-vote rates for the same questions, and learners can explain why the most popular wrong answer is wrong — not just which answer is right.

Masterpiece — the artifact that proves it. A semester-long record of ConcepTest vote distributions (pre- and post-discussion) showing systematic comprehension gains, alongside a curriculum of ConcepTests calibrated to the class's actual misconceptions — a diagnostic and instructional tool in one.

Example scenario

A physics instructor asks: 'A heavy ball and a light ball are dropped from the same height in a vacuum. Which hits the ground first?' After individual think time, the first vote shows 60% chose 'the heavy ball' (a classic Aristotelian misconception) and 40% chose 'they hit at the same time.' The instructor asks learners to turn to a neighbor who answered differently and argue. In 90 seconds, the room is loud. The second vote shows 85% now correctly choose 'they hit at the same time.' The instructor explains — but now the explanation lands differently: every learner who changed their answer has already gone through the specific argument that broke their intuition. The explanation is a confirmation of something they just reasoned through rather than a fact handed to them. The learner who switched from 'heavy ball' to 'same time' because their neighbor argued convincingly is far less likely to revert to the misconception than one who simply heard the right answer stated.

Facets

Container — synchronous
Mode — peer-learningcollaborative-problem-solving
Reach — group
Persona — human-learnerhuman-trainer
Craft (AI Fluency) — discernmentarticulation
Learner — human
Trainer — human

Inputs

ConcepTest — a diagnostic conceptual question — A short multiple-choice question designed to surface a specific known misconception — not a recall test, but a question that requires genuine understanding to answer correctly. The wrong answer choices must reflect the actual wrong intuitions learners typically hold.
Heterogeneous group of learners — A group where some learners have understood the concept and some have not — the productive condition for peer instruction, because the recent understanders are the effective peer teachers.

Outputs

More capable learner — A learner who has actively resolved a specific conceptual confusion through the act of arguing their position with a peer — not just heard the right answer, but worked through why it is right and why the wrong answer is wrong.
Visible learning state — A real-time diagnostic read of where the class is — the masterpiece — surfaced by the vote distributions before and after peer discussion, which tells the instructor which misconceptions are widespread and which are already resolved.

Steps (4)

Pose the ConcepTest and require individual commitment
Present the question and ask every learner to think individually and commit to an answer before any discussion. The individual commitment step is essential: it forces each learner to formulate a position rather than free-riding on a neighbor's answer. Skip this step and the peer discussion becomes a social conformity exercise, not a reasoning one.
First vote — reveal distribution
Collect and display the distribution of answers (by clicker, hand raise, or anonymous poll). The distribution is informative for both instructor and learners: if more than 70% already have the right answer, move on; if fewer than 30% have it, the concept needs direct instruction first; if 30–70% have it, the conditions for productive peer discussion exist.
Peer discussion — turn to your neighbor
Ask learners to find a neighbor who answered differently and try to convince each other. The instruction is to argue, not just share answers: 'Try to convince your neighbor that your answer is correct.' This is the generative step — articulating a position forces the articulator to discover gaps in their own reasoning.
Second vote and explanation
Poll again and compare the distribution before and after discussion. Then explain the correct answer, explicitly addressing the specific wrong intuitions the most popular wrong answers reveal. The explanation now lands differently because learners have already committed to a position and felt the friction of a counter-argument.

Principles

Individual commitment before discussion — without a prior individual answer, peer discussion becomes social information-gathering rather than reasoning. The commitment is what makes the subsequent argument productive.
The recent understander is the more effective teacher — a student who resolved a misconception 10 minutes ago remembers the path through the confusion better than an expert who resolved it years ago. Peer instruction leverages this structural advantage.

Known uses (1)

Eric Mazur — Harvard Physics (1991–present) — Harvard University Department of Physics
course
undergraduate physics education Mazur developed peer instruction after seeing that his Harvard physics students passed standard tests while holding Newtonian misconceptions — as measured by the Force Concept Inventory developed by Halloun and Hestenes (1985).

Known failure modes (3)

[concepTest-as-recall]
The anti-pattern of writing questions that test memory rather than conceptual understanding — questions where the right answer is retrievable without genuine comprehension. A recall question produces no productive peer discussion because the understander and the non-understander both just need to remember, not reason.
[discussion-without-prior-commitment]
The anti-pattern of skipping the individual commitment step and moving directly to peer discussion. Without a prior individual answer, learners adopt a neighbor's position through social conformity rather than updating through reasoning — the mechanism that produces learning is bypassed.
[vote-ignored-after-discussion]
The anti-pattern of not running the second vote after peer discussion, removing the diagnostic read that tells the instructor whether discussion actually moved the class and which misconceptions persist.

Related trainings (4)

Sources (2)

Provenance

Ecosystem: general education
Added to catalog: 2026-05-27
Last updated: 2026-05-27
Verification status: verified

How the learner advances

Steps (4)

Pose the ConcepTest and require individual commitment

First vote — reveal distribution

Peer discussion — turn to your neighbor

Second vote and explanation