Cognitive Apprenticeship

Pattern · All patterns
Evidence · 28 claims (12 for, 13 mixed, 3 against) · 26 studies (9 causal, 5 quant-synthesis, 4 review, 4 qualitative, 2 associational, 1 design, 1 theoretical), q1–q4 · 3 of 26 report an effect size · 24 claims rest on one study

Description and scope

This is a reusable policy for teaching a complex cognitive skill whose expert reasoning is not visible in the finished product: writing, diagnosing, debugging, reading for meaning, solving a non-routine problem. An expert makes the reasoning visible (modeling). The learner attempts the task while the expert watches and responds (coaching), and gets support matched to where they struggle (scaffolding). That support is withdrawn as the learner shows they can work without it (fading). The learner then states their reasoning and compares it with the expert's (articulation, reflection), and finally applies the skill to new problems with little guidance (exploration). It instantiates the scaffolding and fading principle. The study configurations below are evidence about parts of this sequence. The whole sequence, and the response-dependent policy that moves a learner between its phases, is an untested design proposal: no claim in this wiki tests the full six-phase arc against an alternative.

Inputs: establish the design brief

Field	Record or elicit
Learner and starting observation	Task-specific experience; an unassisted attempt at a representative task, with the learner's reasoning, errors and confidence. Preserve unknowns. A professional or graduate learner can still be a novice for this task.
Context and activities	Setting (one-to-one, small group, whole class, online-self-paced, workplace-clinical); who models (expert, peer, recorded); time available; how coaching and support are delivered and logged; tools and aids permitted.
Learner-valued goal	Ask what the learner wants to be able to do and why. Record it apart from the designer's objective; agreement is an observation.
Designer objective	The capability (complex-skill, procedure, metacognitive-strategy), a representative task, what must be done without support, and which aids remain legitimate at the end.
Outcome	Instrument and local criterion; immediate and delayed horizons; whether a standardised or researcher-made measure is used (they give different effect sizes, see below); any real-use outcome.

A request to "use cognitive apprenticeship" does not specify these. Ask for them before fixing the length of each phase, a fading schedule or an expected gain.

Sequence and conditional policy

1. Elicit. Ask for an unassisted attempt and the learner's reasoning before any modeling. Keep it short; it is a probe, not a treatment. Record it.
2. Model with reasons. Show expert performance on a representative task, narrating the decisions and the cues that triggered them, not just the steps. If the learner already produces adequate unaided reasoning on varied cases, shorten or skip modeling (expertise reversal).
3. Coach and scaffold contingently. The learner attempts a comparable task. Raise support after a failure and lower it after a success, starting from the least support that might do (a cue before a hint, a hint before a partial solution). Record each level given and whether the learner applied, copied or ignored it.
4. Check before withdrawing. Before removing a support, ask the learner to do or explain the next step without it. Withdraw gradually, and announce it.
5. Articulate and reflect. Ask the learner to state their reasoning and compare their process with the model, naming one difference that mattered.
6. Explore and reobserve. Pose a changed problem with minimal guidance, under the agreed aids and at the stated horizon. If the response disagrees with the expectation, revise the interpretation or the configuration, not only the learner's label.

Observation	Discriminating probe	Proposed action, with its uncertainty
Cannot start after watching the model	Ask what the expert did first and why; give a matched task with a first-step cue only.	If the learner can name the reason but not act on it, coach the first step with minimal prompts. If they cannot name it, model again on a contrasting case, with the decision point marked. Neither branch is a validated diagnosis.
Succeeds only at the support level last given	Present the next comparable step at one level lower (a cue instead of a hint).	If the lower level is enough, keep fading one level at a time. If not, hold the level and change the case. One success does not establish readiness; no fading threshold is supported.
Correct answer, but copied rather than applied	Ask for an explanation, and pose a changed case the copied move would not solve.	Treat the support as not yet taken up. Check understanding explicitly before the next withdrawal; the link to later accuracy is associational.
Fluent with the expert's model, poor on a changed problem	Vary one relevant condition, not only surface details; ask what changes and why.	Add articulation and comparison across contrasting cases before exploration; do not repeat the same model.
Avoids or refuses support it still needs; or relies on support it no longer needs	Offer the task with support optional; ask privately about purpose and barriers.	Negotiate the goal or conditions when they drive the pattern. Help use alone is not evidence of competence or of its absence.

Choosing configurations from evidence

• The whole approach. Cognitive apprenticeship methods were reported more effective than traditional methods for college writing [+W]. This is second-hand: a narrative review reports an earlier study's finding, with no design detail and no effect size. The judge checked the review's text, not the original study. Treat it as a pointer, not as a forecast.
• Modeling with gradual handover, in reading. Reciprocal teaching improves reading comprehension [+M]. Teachers model four strategies and hand the discussion-leader role to students. A review of 16 studies reports a median effect of .32 on standardised comprehension tests and .88 on researcher-made tests. The outcome instrument changes the size of the benefit, so state which one a design will use. Read from an abstract; not yet checked against its sources.
• Model first, then attempt. Example–problem sequences reduce cognitive load and improve learning outcomes [+M]. Dutch secondary-school novice learners in circuit troubleshooting (96 analysed, randomised) did better on an immediate test, with lower reported mental effort, when training began with a worked example than when it began with a problem. This supports putting modeling before attempts in a short procedure task. It establishes no delayed or far-transfer effect. Not yet checked against its sources.
• Contingent coaching. Contingent scaffolding improves learning more than fixed or absent support. [+M]. In small one-to-one tutoring studies, fully contingent support beat fixed or no support in long division, at an immediate and a one-month test. A dynamic-assessment synthesis ranked explicit strategy training above scaffolding and scaffolding above coaching. So "coaching" without contingent support, or in place of explicit strategy teaching, is not what this evidence supports. Not settled: abstracts could not confirm the entries.
• Fading. Fading support promotes the transfer of responsibility from instructor to learner. [+M]. Across three experiments, a gradual move from complete to incomplete examples to independent problems favoured near transfer over static example–problem arrangements, and removing the last steps first did better than removing the first steps first. No fading rate is established. Not settled: abstracts could not confirm the entries.
• When to shorten the arc. Worked-example guidance becomes less effective as learner expertise increases. [~M]. Full guidance helps novice learners more than knowledgeable ones and can become redundant for them, so a learner who already performs unaided should not be sent through every phase. Reviews only, not settled against their sources. They give no individual cut-off.

Do not rank these against one another or average them: their learners, tasks, comparators and outcomes differ. Nothing here supports a fixed duration for each phase (the old "weeks, not a single lesson" is a design judgement, not a finding).

The evidence by question

Each question below links a general claim that gathers the studies behind it into one table, one row per distinct study, with each study's design and effect as recorded and no averaged verdict (eval/synthesis/cognitive-apprenticeship.json, scripts/synthesize_claim.py). The count is of distinct studies across the claim and the narrower claims that bear on it.

• Does the whole cognitive-apprenticeship approach beat traditional instruction? The review reports that cognitive apprenticeship instructional methods were significantly more effective than traditional methods for college-level writing skills [+M]: 4 distinct studies.
• Does modelling strategies and handing the lead to learners improve reading comprehension, and on which measures? Reciprocal Teaching Improves Reading Comprehension [+M]: 2 distinct studies.
• Should modelling come before the learner's first attempts? Example-based sequences outperform problem-only practice for novices, and fading support is argued to aid transfer [+S]: 7 distinct studies.

Fitting the design to a situation

Ask for three facts when the brief omits them: the learner's unassisted attempt and reasoning on a representative task, who coaches, and how many learners each coach serves, and which outcome measure and horizon will judge the design.

If the brief says…	Then change…	Basis
Learners are new to this task	Put modelling first: a worked or narrated example, then the learner's attempt, rather than starting with a problem	Example-problem sequences reduce cognitive load [+M]; Example-based sequences outperform problem-only practice [+M]
The target is reading comprehension	Use reciprocal teaching: model summarising, questioning, clarifying and predicting, then hand the discussion lead to learners	Reciprocal teaching improves reading comprehension [+M]; Reciprocal teaching raises comprehension of poor readers [+W]
Success will be judged on a standardised test	Expect a smaller gain than researcher-made tests show (median .32 against .88 for reciprocal teaching) and set expectations accordingly	Reciprocal teaching improves reading comprehension [~M]
One-to-one or small-group coaching is possible	Coach contingently: least help first, more after a failure, less after a success	Contingent scaffolding improves learning [+M]
One instructor coaches a whole online cohort	Set limits on coaching time, and use recorded models or peer coaching for part of it; coaching without a time limit became a large, open-ended load for the instructor	Web-based cognitive apprenticeship required large instructor time [~W]
Support moves from worked examples to independent problems	Fade from complete to incomplete examples to independent problems across the cases, not in one step	Fading support promotes transfer of responsibility [+M]
A learner already performs the task unaided	Shorten or skip modelling and coaching for that learner and move to articulation and exploration	Worked-example guidance becomes less effective with expertise [~M]
The course replaces traditional instruction in technical mathematics and is judged on exams	Do not promise exam gains; one community-college quasi-experiment found none, and higher mathematics anxiety alongside more confidence. Track anxiety as an outcome	No significant exam differences in technical mathematics [-M]; Anxiety rose while confidence grew [~M]
Learners are learning to program or debug	Order tasks from completing a program to fixing a faulty one to building one; reasoning shifted toward diagnosis as tasks moved on	Scaffolded task progression shifts novices toward diagnosis [+W]
The brief fixes a length for each phase (weeks of modelling, then coaching)	Move learners between phases on their unaided performance instead; no phase length is established	untested proposal

Further evidence, not yet read against this model

Claims this page cited before it was rewritten as a conditional model. They are evidence about the relationship, but none has yet been re-read against the model above; each says how far its own sources have been checked.

• Worked examples reduce unnecessary search for novices. [+M] — not settled: the text available could not confirm the entries (abstract)
• Example-based sequences outperform problem-only practice for novices, and fading support is argued to aid transfer [+S] — not settled: the text available could not confirm the entries (abstract)
• Mentored inquiry with staged mechanistic sensemaking supports the transition from intuitive to formal equation reasoning [+W]
• A network of three mutually overlapping intuitive ideas (less distance implies less time, bunched up means faster, faster implies less time) stabilizes students' initial thinking [~W]

Illustrative design instance and observation record

Illustration only, not a tested design. A first-year nursing student wants to stop freezing when a patient's vital signs change. The designer wants the student to reason aloud from observations to a priority action. Agree that both matter, and that the target is a simulated deterioration case handled without prompts, assessed with an agreed reasoning rubric now and again in four weeks. The instructor models one case, narrating which cue changed their priority. The student then works a second case with prompts that are withdrawn after the student explains the next step correctly. The student articulates their reasoning against the instructor's, and finally works a changed case alone. None of these choices (the number of cases, the four-week horizon, the withdrawal rule) is established by the evidence above.

Record: task and conditions → model shown (what was narrated) → learner attempt → support level given and whether it was applied, copied or ignored → check before withdrawal and its result → candidate interpretations and their basis → next activity and why → unassisted response at the stated horizon. Keep the learner's stated purpose and any change in it. An agent may summarise this record, but must not invent missing inputs or assign numerical probabilities to a learner's state without a calibrated model.

Elements and limits

Modeling, think-aloud, demonstration, worked examples, coaching, eliciting student thinking, feedback, hints, scaffolding, fading, articulation, reflection and practice. Principles the sequence draws on: worked examples, scaffolding, guided practice and purposeful reflection.

This pattern is general across domains; a clinical round, a writing workshop or a design studio is a design that uses it, not a separate pattern. It depends on someone being able to make their reasoning explicit, and contingent coaching is hard to sustain for many learners at once. The evidence for contingency comes mostly from one-to-one tutoring. Whether the full sequence improves independent performance more than explicit strategy instruction, or more than its parts used alone, has not been tested here, and neither has its diagnostic policy.

Claims

• Scaffolded task progression shifts novice reasoning from local completion toward diagnosis and systematic debugging as task demands increase [~W]
• Integrating cognitive apprenticeship methods in a Web-based course required a significant and largely unbounded instructor time investment [~W]
• Scaffolding effects varied by education population, with traditional learners significantly outperforming underperforming learners [~W]
• No effect of fading on productive learning events including self-explanations was found in Experiment 2's think-aloud analysis [-W]
• Teacher and researcher coaches both used a cognitive apprenticeship approach but differed in scaffolding style [~W]
• Expert scaffolding is associated with ILS adoption, but the fading stage shows a comparatively low rate (8.27%) and was a non-significant predictor [~W]
• Model-observer gender combination affects mental effort invested during example study: male students invest less effort with a male model, female students more effort with a male model [~W]
• No significant differences in learning measures across teacher-coached, researcher-coached, and BioWorld-alone conditions [-W]
• Students start with video modeling examples but shift toward worked examples and practice problems as the learning phase progresses [~W]
• Scaffold effects peak at 2–4 weeks of treatment and decrease dramatically for treatments lasting 9–24 weeks [~W]
• Experts introduce procedure and planning explanations over time, typically after an initial step or action is completed [+W]
• Experts scaffold by reassuring novices to continue acting or thinking, removing uncertainty about whether they chose the right action [~W]
• Active verbal cues from experts most often prompt nonverbal action from novices, often preceding explanatory information [~W]
• Customized scaffolding (fading/adding) outperforms non-customized scaffolding in PBL-STEM, with fading/adding highest at g = 0.590 [~W]
• Expert modeling, feedback, hints, and multiple strategies improve cognitive outcomes, but question prompts show no evident difference from control [+W]

Related Patterns

• Four-Component Instructional Design — shares the principle of worked examples fading to full tasks; provides more formal scaffolding design rules for complex learning
• Guided Discovery Learning — similar progression from support to independence, but begins with learner exploration rather than expert modeling
• Three fading-based scaffolding approaches: supportive, intrinsic and reflective scaffolding
• (none yet linked)

Examples

Medical education — clinical rounds: Attending physicians model diagnostic reasoning by thinking aloud through a patient case, then coach residents through their own case assessments with progressively less guidance over weeks.

Writing instruction — writer's workshop: Teacher models drafting and revision strategies on a shared text, thinking aloud about audience and structure; students write alongside with peer and teacher coaching; feedback is gradually reduced as writers develop independent revision habits. Uses teacher modeling, conferencing, and author's chair sharing.

Engineering education — design studios: Expert designers walk through a design decision process on a real project; students work on their own projects with structured critiques (charettes) that fade from expert-led to peer-led over the semester.

Replit and paired programming environments: Expert-novice pairing where the expert narrates code decisions; over time the novice takes the keyboard while the expert coaches. Fading occurs as the novice's contributions increase. - Interactive Modeling - Model Empathy and Explain - Connecting Struggles to Strategies - Mentor Text Analysis - A high-school student researching wave velocity in liquids spent under 10 minutes deriving the phase-velocity equation v = (gλ/2π + σ/2πλρ)tanh(2πh/λ), but "hours and days" across three mentoring sessions working through what each term meant, why surface tension was negligible in his experimental conditions, and why velocity increased with depth — moving over the course of the year from "classic ADHD" disengagement (his 10th-grade teacher's description) to sustained, self-directed inquiry.

Key Sources

• Collins, A., Brown, J. S., & Newman, S. E. (1989). Cognitive apprenticeship: Teaching the crafts of reading, writing, and mathematics. In L. B. Resnick (Ed.), Knowing, learning, and instruction: Essays in honor of Robert Glaser (pp. 453–494). Lawrence Erlbaum. doi:10.4324/9781315044408-14
• Collins, A., Brown, J. S., & Holum, A. (1991). Cognitive apprenticeship: Making thinking visible. American Educator, 15(3), 6–11, 38–46.
• Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge University Press. doi:10.1017/cbo9780511815355
• van Merriënboer, J. J. G., & Kirschner, P. A. (2018). Ten steps to complex learning (3rd ed.). Routledge. doi:10.4324/9781315113210
• Dozier, C. L. (2008). Literacy coaching: Engaging and learning with teachers. The Language and Literacy Spectrum, 18. https://eric.ed.gov/?id=EJ1059644
• Collins, Allan. (1988). Cognitive Apprenticeship and Instructional Technology. Technical Report. BBN Report No. 6899, BBN Labs, Inc. https://eric.ed.gov/?id=ED331465
• Kapon, S., & Schvartzer, M. (2024). Guided Inquiry into a Physics Equation. Cognition and Instruction, 42(1), 159-206. https://doi.org/10.1080/07370008.2023.2197232

Page metadata

Field	Value
type	pattern
id	cognitive-apprenticeship
aliases	four-reflection-methods-reification, guided-equation-appropriation
title	Cognitive Apprenticeship
description	A reusable modeling, coaching, scaffolding, fading, articulation and exploration policy for complex cognitive skills, whose phase transitions depend on observed unaided performance and remain an untested design proposal.
status	review
grain_size	course, unit
author	Collins, Brown, & Newman (1989)
moved_from	principle/notice-and-name-instructional-practices
canonical	True
generated	by claude/unspecified on 2026-10-01
trust	unverified — no one has confirmed this page's content
sources	5 — lave-1991, van-merriënboer-2018, dozier-2008, collins-1988, kapon-schvartzer-2024

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