Skip to content

Modeling

Principle · All principles
Evidence · 30 claims (23 for, 7 mixed) · 43 studies (13 causal, 10 review, 5 quant-synthesis, 5 qualitative, 4 theoretical, 3 associational, 3 design), q1–q4 · 4 of 43 report an effect size · 20 claims rest on one study

Conditional relationship

This page owns making the process of a performance visible before and alongside the learner's own attempts: showing the decisions behind a piece of work (what the performer notices, chooses, checks and corrects, and why) rather than only the finished product. For a learner whose first response on a task shows they cannot yet produce the performance, or can produce a surface imitation of the product without its decisions (a novice for this task, on a procedure, principle, complex-skill or metacognitive-strategy whose key moves are not visible in the product), a model that externalises those decisions, studied actively and followed at once by the learner's own supported attempt, is expected to produce better unaided performance on comparable tasks (immediate-performance, near-transfer) than either the finished product alone, a silent demonstration, or being left to work out the process unaided. Watching alone is not the expected active ingredient: the expectation is stronger when the learner has to explain or predict the modelled decisions, and it weakens as the learner's task-specific expertise grows. Whether the gain holds at delayed-retention or far-transfer is not settled here.

No claim in the wiki tests this relationship as stated. No recorded study compares a narrated (think-aloud) model with a silent demonstration or with the product alone, or varies the kind of model (expert, peer, a model who makes and repairs mistakes). The nearest tested evidence is about worked examples, a written form of modelling: studying solved examples before solving beats solving alone for novices, prompting learners to explain what they study helps on immediate tests, and the advantage of full examples shrinks with expertise. Claims about live modelling itself are single studies, mostly qualitative or design reports. The model below carries the worked-example findings to live, spoken and video modelling, and says so wherever it does.

How the converted neighbours sit beside this one: Worked Examples owns studying a written solution before a comparable problem, and is the best-evidenced special case of modelling. Direct Instruction owns the whole explain–model–guide–hand-over sequence and its comparison with discovery; modelling is one of its moves, treated here on its own and for tasks (writing, discussion, professional conduct, play, motor and social skills) that have no written solution. Guided Practice owns the attempts that follow the model, and Scaffolding and Fading owns how support is withdrawn; this page hands over to both and does not repeat their models. The Cognitive Apprenticeship pattern is a reusable policy that opens with modelling; its branches are untested proposals.

Default design, while the relationship is untested

The earlier page's guidance (a visible process, narrated reasoning, an immediate link to practice; slow down, segment and annotate for novices; reduce explicit modelling as expertise grows), kept as a concrete default a designer can act on and revise. Each step says what evidence it rests on.

  1. Find the decisions the product hides. Before modelling, list the three to six decision points where a novice's attempt goes wrong or stalls (what to notice first, which option to choose, what to check, when to stop and revise). Model those, not every keystroke. Untested proposal; from the earlier page ("what to notice, what to prioritize, why a step happens now instead of later, and how errors are handled").
  2. For a learner new to the task, model before the first attempt. Show the model first, then the attempt; do not open with an unaided attempt and use the model as feedback. Evidence carried from written worked examples: example-first sequences beat problem-first ones and problems alone for novices on an immediate test (Example-based sequences outperform problem-only practice for novices, and fading support is argued to aid transfer [+M]; Worked examples reduce unnecessary search for novices. [+M]). Learners with some experience are different (step 10 and the situation table).
  3. Narrate the reasoning, not only the action. Think aloud at each decision point: "I'm checking X first because…", "I could do A or B; I choose B because…". Keep each narrated segment short: about 3–5 minutes, covering one to three decision points, then pause. The narration and the segment length are untested proposals; from the earlier page's warning that silent demonstration hides the cognition and expert performance moves too fast for novices.
  4. Make the learner do something with each segment. At each pause, ask learners to predict the next move, or to say (or write in a line) why the move just made was made. Evidence: prompting learners to explain what they study helps on immediate tests by a moderate average amount, across task types (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. [+S]); constructive activity beating merely watching is supported in the lab but was not confirmed in one large university field study (Constructive learning beats active and passive learning [~M]).
  5. Include a mistake and its repair. In at least one segment, make (or show) a typical error, notice it aloud, and fix it, then ask learners what gave it away. Evidence carried from tutoring with fourth and fifth graders on decimals: comparing correct with incorrect examples beat comparing correct examples only (Erroneous examples improve conceptual understanding by forcing comparison with correct models. [+M]); untested for live modelling.
  6. Show a second, different instance. Model the same decisions on a second task whose surface differs, so learners can see what stays the same; for open tasks (writing, design), show two acceptable products and name what they share. Untested proposal; a single learner comparing his writing to one model paragraph noticed only vocabulary (Comparing original texts to model paragraphs prompted noticing that was again focused on lexical features [~W]).
  7. Hand over at once. Follow each model within the same session with the learner's own attempt on a comparable item, with help still available, then an unaided item. Evidence: example-first sequences that included problems did as well as examples only and better than problems only on an immediate test (Example-based sequences outperform problem-only practice for novices, and fading support is argued to aid transfer [+M]); how to run the attempts is in Guided Practice.
  8. Re-model at the decision the errors show. When several learners make the same error, model again only that decision, narrating what you check; do not repeat the whole demonstration. Untested proposal.
  9. Plan the withdrawal of the model across sessions. A default arc for a unit of four to six sessions: sessions 1–2, a full narrated model of a whole task before practice; sessions 3–4, a partial model (the teacher starts, learners finish and narrate the rest), with the full model kept available for reference; sessions 5–6, learners model for each other (one narrates while a partner checks against a criterion list), and the teacher models only at decision points the errors show. Move a learner to the next stage once they can do the modelled decisions unaided, and say why, on two consecutive items. Evidence for fading from complete to incomplete examples, on near transfer (Fading support promotes the transfer of responsibility from instructor to learner. [+M]); the arc, the session counts and the threshold are untested proposals. How to fade is modelled in Scaffolding and Fading.
  10. Keep full models for those who need them; shorten them for those who do not. Learners with low prior knowledge on this task keep complete models longer; learners who succeed on the first unaided item get a brief model of only the new decision, or attempt first and then compare with the model. Evidence: guidance that helps novices can become redundant as expertise grows (Worked-example guidance becomes less effective as learner expertise increases. [~M]; Instructional guidance that helps novices can become redundant or counterproductive as expertise grows. [~M]); in one algebra experiment with college students, lower-prior-knowledge learners did better after full than completion examples (Lower-prior-knowledge learners scored higher on an algebra posttest after full-worked than completion-worked examples, while higher-prior-knowledge learners' non-significant advantage ran the other way [~M]). Neither gives a crossover point.

When a condition is not met: if no one can watch learners' attempts after the model (self-paced video), embed the predict-or-explain pauses in the recording and follow it with automatically checked items; if the task has no hidden decisions (the product shows them all), a worked product with annotations may be enough and live modelling adds time without a recorded gain.

Fitting the design to a situation

The three facts that most change the decision:

  • What the learner can already do on this task. A learner who cannot begin needs a full, slowed, narrated model first; one who succeeds unaided needs at most a brief model of the new decision, and a full model may be redundant. If unstated, ask: can they produce one example of this performance now, unaided, and say why they made its main choices?
  • Which decisions the finished product hides. If the product shows every step (a solved equation), an annotated product may do; if the decisions are invisible (why a writer cut a paragraph, why a nurse checked a line first, how a mentor defused a classroom incident), they have to be narrated or they are not modelled. If unstated, ask: where do novices' attempts go wrong, and can that choice be seen in the finished work?
  • What the model is for: the performance, or the learner's belief they can do it. Accurate performance calls for an expert's clear, narrated model; confidence and persistence may call for a model the learner sees as similar to themselves, who struggles and recovers. If unstated, ask: is the problem that learners cannot do it, or that they do not try?
If the brief says… Then change… Basis
Learners are advanced on this task, or succeed on the first item unaided Skip the full model: let them attempt first, then compare their work with an expert model and name the differences; model only the one new decision Instructional guidance that helps novices can become redundant or counterproductive as expertise grows. [~M] (review of cognitive-load studies; no crossover point)
Learners have very low prior knowledge of the task Keep complete models (every step shown and narrated) for longer before any partial or completion model; slow down and segment further (one decision per pause) Lower-prior-knowledge learners scored higher on an algebra posttest after full-worked than completion-worked examples, while higher-prior-knowledge learners' non-significant advantage ran the other way [~M] (college algebra, one experiment; carried to live modelling)
Learners are young children, and the target is play, social or motor behaviour Model by acting out with materials in small groups, over several short sessions (the study used eight), and let children re-enact straight after Modeling training improved imaginative play in disadvantaged kindergartners, with gains in imaginativeness, verbal communication, spontaneity, and attention span persisting two months [+W] (80 kindergartners, a 1973 report read second-hand; modelling combined with direct teaching)
Goal is a concept or principle, or transfer, not only a procedure Model on two items that differ on the surface; at each pause ask why, not what; include one error and its repair, and ask learners what gave it away 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. [+S]; Erroneous examples improve conceptual understanding by forcing comparison with correct models. [+M] (grades 4–5, decimals, tutoring; carried to live modelling)
Goal is confidence, persistence or willingness to try (attitude-motivation), not only skill Use a model the learners see as like themselves (a peer, a former student, the teacher showing an early struggle), who makes and recovers from mistakes and says what they did next; pair it with an early task they can succeed at Learners' task-specific self-efficacy beliefs predict whether they initiate, persist through, and recover from difficulty on academic tasks. [+M] (two reviews; names watching similar others as one source of self-efficacy, does not compare model types). Proximal goals and ability attributional feedback for early learning raise self-efficacy more than distant goals and effort feedback, by a review's report of earlier studies [~W]: its title names peer models, but neither of its entries reports them
adult learners in a work setting (workplace-clinical), or teachers in training Model the performance in the real setting (the mentor's own lesson, a supervised shift), then debrief the decisions it involved, then have the novice try the same move with the mentor watching Mentor modelling and a school-wide behaviour program's strategies transferred to the mentee's classroom management practices [+W] (one preservice teacher, qualitative case); Vicarious experiences are positively related to agriculture teachers' general teaching efficacy, with observing a first year teacher explaining 11% of variance [+W] (correlational, reported in a review)
Writing, composing or other open tasks with no single right answer Draft live in front of learners, thinking aloud about audience and choices, and revise visibly; show two different good products and name the features they share, so learners do not copy one product's wording Comparing original texts to model paragraphs prompted noticing that was again focused on lexical features [~W] (one EFL learner, case study); Teacher modeling with worked examples helped EFL students internalize reading comprehension strategies [+W] (students' questionnaire comments, one study); otherwise untested
Pronunciation, music or other performances learners can record Have learners record their own attempt, listen to it, then hear the model, then try again Students are more attentive to the teacher's model track when they hear their own recording first [+W] (language lab, design report; attention, not learning, was observed). For true novices on a new performance, model first (step 2 of the default)
Setting is online-self-paced, or video only Record short clips, one decision each (about 2–5 minutes), each ending in a predict-or-explain question the learner must answer before the next clip; follow with automatically checked practice items untested proposal; the explanation prompt is carried from Prompting learners to self-explain… [+S]
Some learners are remote while others are in the room, or some join by phone only Model on a shared screen or document camera rather than a whiteboard; narrate into the microphone; at each pause collect predictions in the chat or a poll from everyone, remote first; record the model and share it after for replay. For phone-only learners, send the recording cut into short clips with the pause questions in text untested proposal
One teacher with a large group Give one live model to all; at each pause have every learner show a prediction at once (cards, mini-whiteboards, a poll); then in pairs, one learner narrates the next item while the other checks against the modelled decision list; re-model only for the error most pairs show untested proposal; the explanation step is carried from Prompting learners to self-explain… [+S], and the large field study under Constructive learning beats active and passive learning [~M] found only the most interactive activities ahead of passive ones in real classes
Time is one session only Model one item fully with narration, then a partial model the learners finish, then one unaided item; record what the unaided item shows and do not claim retention Example-based sequences outperform problem-only practice for novices, and fading support is argued to aid transfer [+M] (example-first beat problem-first in one short troubleshooting task, immediate test)
Learners have low literacy, or work in a second language Model by doing more than by telling: fewer words, gestures, labelled pictures; name and display only the key terms; let learners show the next move rather than explain it in words untested proposal
The performance is for a real audience, or a product whose audience should not see the reasoning (a client-facing call, a published piece) Model the finished performance once at full speed as the audience would see it; then model it again slowed down with the reasoning narrated, in a rehearsal space the audience never sees; keep the reasoning in a debrief or a separate annotated version untested proposal
High stakes or safety (clinical, lab, machinery) Model the whole procedure once at normal speed, then step by step with what to check and why; include the commonest error and how it is caught; require an unaided demonstration under supervision before the learner works alone untested proposal; the error step is carried from Erroneous examples improve conceptual understanding by forcing comparison with correct models. [+M]

Twelve of the fifteen rows cite a claim, but none of those claims tests live, narrated modelling against an alternative. Three (self-paced video, one teacher with a large group, high stakes) are untested proposals that carry one component from a claim; four carry a worked-example, self-explanation or erroneous-example finding to live modelling; the rest rest on single studies, a review that names modelling as one source of self-efficacy, a case study or a design report, as each row says. The mixed-mode, low-literacy and real-audience rows are untested proposals.

Observation, state and explanation

What can be observed is what the learner does and says after a model under recorded conditions: which model they saw (live or recorded, narrated or silent, expert or peer, with or without an error), what they predicted or explained at each pause, their next attempt, and whether that attempt reproduces the model's decisions or only its surface. "Has understood the process" and "is copying" are inferred states. A correct imitation right after the model shows what the learner can do with the model fresh in mind; it does not show what they can do on a different task or a week later.

Response Plausible explanations Observation that could change the interpretation
The learner's product closely copies the model's wording or layout, but the decisions are wrong on a new task Copying the surface; only one model seen, so its features and its decisions cannot be told apart; the decisions were not narrated, or not attended to Give a task where the model's surface would be wrong, and ask the learner to say which of the model's choices they kept and why.
Watched attentively, predicted well at the pauses, then cannot start alone The prediction relied on the model being in view; the first decision was never modelled (the model began after it); low confidence rather than missing skill Ask for the first move only, with the model hidden, then with only the decision list; ask how sure they are before they start.
Can perform the steps but cannot say why A procedure learned without its reasons; a language barrier to explaining; the narration gave actions, not reasons Pose an item where the usual step is wrong, and allow a non-verbal explanation (point, draw, act).
Finds the model too slow, or tunes out The learner already has the decisions (redundancy); the segment is too long; the task or model is not one they value Give one unaided item at once: if it is right, shorten or skip the model; if wrong, keep the model and shorten its segments.
Avoids trying after an expert model, though the attempts show partial skill The expert's ease made the task look out of reach; fear of being seen to fail; the learner does not see themselves in the model Show a peer or coping model of the same task, or the teacher's own early attempt, and compare willingness and first-attempt quality.
Performs well right after the model, much worse a week later Little consolidation; the attempts were too close to the model; no practice between Re-test with a new item at the delay, and record what practice and exposure happened in between.

These rows are proposals, untested with learners as diagnostics. A probe is itself practice and changes what it measures, so record each one, and keep open the explanations that remain rather than choosing the one whose name fits.

Evidence and qualifications

No claim in the wiki compares narrated with silent modelling, process with product, or one kind of model with another. The claims below test neighbouring relationships, mostly through written worked examples.

Learners, tasks and outcomes differ across these claims (algebra novices, circuit troubleshooters, children with a tutor, university classes), so do not rank them by effect label. None measures a modelled skill at delayed-retention against a condition without the model.

Single studies of modelling itself, each weak and cited in the situation table only for the setting it observed:

The evidence by question

Further evidence, not yet read against this model

Claims that mention modelling but bear on this page little or not at all:

Objective and learner-valued goal

The designer's objective is usually unaided performance on a representative task, with stated aids, at a stated horizon, and a model is a means to it. Elicit separately what the learner wants from watching: to see that the task is possible for someone like them, to get a template they can copy for tomorrow's assignment, to understand how an expert thinks, or simply to get the session over. Agreement and divergence are both observations. A learner who wants a template will attend to the model's surface and copy it, which serves a deadline and not the objective; one who doubts they can do the task may need a model they identify with before a clearer expert model helps.

For example, a new nurse may want above all not to look incompetent in front of a patient, while the designer's objective is a correct sequence of checks before giving a medication. A model at full speed by a senior nurse shows what competence looks like to the patient; a slowed, narrated rehearsal of the checks, away from the patient, serves the objective. Record both, so that a calm performance on the ward is not read as the checks having been learned, or the reverse.

What would revise this model?

The expectation should weaken if comparisons of narrated with silent modelling, or of a process model with an annotated product, with comparable novices and aligned unaided outcomes, show no advantage, or an advantage only at the end of the session. If asking learners to predict or explain during a model works no better in real classes than watching, as one large field study of activity modes suggests for constructive activity in general, the pause prompts are a lab finding only. If peer or coping models raise performance or persistence no more than expert models, drop the model-choice row. If learners who attempt first and then study a model do as well as those modelled first, even as novices, the order in step 2 should be relaxed. Do not protect the model by calling every failure "the model was too fast" or "not enough narration" after the fact.

Imitation right after the model, unaided immediate performance, delayed retention, transfer, confidence and use in the setting the learner values are separate claims. The present evidence establishes no segment length, no number of models, no kind of model and no point at which to stop modelling.

Claims

  • Observation/Shadowing — both principles rely on seeing performance, but modeling is usually more intentionally annotated and instructional
  • Explicit Instruction — modeling is one of the core mechanisms that makes explicit teaching actually explicit
  • Cognitive Apprenticeship — extends modeling into coaching, scaffolding, and fading
  • Worked Examples — written solutions studied before a comparable problem; the best-evidenced form of modelling
  • Guided Practice — the supported attempts that should follow a model at once
  • Self-Explanation — prompting learners to explain what they watch or study

Examples

Illustrative

Teacher Modeling — An instructor drafts a paragraph live, narrating audience decisions, sentence choices, and revisions so learners see writing as a process rather than a finished artifact.

Explicit Teacher Modeling — A teacher demonstrates a procedural skill step by step, stopping to name critical features, common errors, and decision points before students try it.

Cognitive Apprenticeship — Modeling serves as the entry point into a larger sequence of coaching, scaffolding, articulation, and fading.

Think-Aloud Modeling, Modeled Writing and Video Based Modeling — strategy recipes for narrated, written and recorded models.

Key Sources

  • Bandura, A. (1986). Social foundations of thought and action. Prentice Hall.
  • Collins, A., Brown, J. S., & Holum, A. (1991). Cognitive apprenticeship: Making thinking visible. American Educator, 15(3), 6–11, 38–46.
  • Scharlott, L. J.; Rippey, D. W.; Rosa, V.; Becker, N. M. (2024). Progression toward Causal Mechanistic Reasoning through Phenomenon-Based Learning in Introductory Chemistry. Journal of Chemical Education. https://doi.org/10.1021/acs.jchemed.3c00517
  • Merrill, M. D. (2002). First principles of instruction. Educational Technology Research and Development, 50(3), 43-59. https://doi.org/10.1007/BF02505024
Page metadata
Field Value
type principle
id modeling
aliases model-cmr-through-observable-phenomena
title Modeling
description For a novice on a task whose key decisions a finished product hides, a model that narrates those decisions, studied actively and followed at once by supported practice, is expected to give better unaided performance than the product alone or no model, though no claim here tests live or narrated modelling directly; the evidence is carried from worked examples and self-explanation.
status review
moved_from pattern/merrills-first-principles-demonstration
canonical True
generated by claude/unspecified on 2026-10-05
trust unverified — no one has confirmed this page's content
sources 2 — scharlott-2024, merrill-2002

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