Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Guided Instruction, Scaffolding, and Fading

Abstract. Guided instruction, scaffolding, and fading describe an instructional system in which assistance is aligned with a learner’s developing competence and progressively reorganized toward independent performance. This fundamental research monograph integrates the historical origins of scaffolding, sociocultural accounts of assisted performance, cognitive-load explanations of worked examples, classroom research on teacher contingency, inquiry-learning meta-analyses, studies of fading instructional prompts, and contemporary computer-based and artificial tutoring. Evidence is examined through constructs, research designs, transfer criteria, heterogeneity, and the distinction between immediate success and retained competence. The synthesis establishes substantial evidence that structured guidance can improve learning while identifying unresolved questions about how to diagnose readiness, schedule assistance withdrawal, preserve productive challenge, and extend effects across subjects, student populations, and real classrooms.

Keywords: guided instruction; scaffolding; fading; contingent teaching; worked examples; expertise reversal; teacher–student interaction; instructional guidance; independent performance; transfer of learning; cognitive load; adaptive tutoring

The Scientific Object of Guided Instruction

Guided instruction is the deliberate organization of instructional actions through which a more knowledgeable person or designed system assists a learner in acquiring knowledge and skills that are not yet reliably available for independent use. The construct concerns a sequence of interactions: assessment of current performance, selection of informative support, learner participation, observation of the response, and adjustment of subsequent assistance. A guided lesson may include explanation, modeling, prompts, worked examples, targeted questions, feedback, or constrained practice. These methods become an instructional system when their timing and content respond to the knowledge being acquired. The research question is therefore more specific than whether teachers explain material: it asks how different forms and doses of guidance change independently demonstrable learning. The knowledge-learning-instruction framework helps specify both the knowledge component targeted by teaching and the learning event expected to alter it Koedinger et al. (2012).

The empirical outcome of interest is a change in what learners can do, explain, retain, or transfer after assistance changes or disappears. A learner who copies a modeled solution accurately may be displaying successful participation without possessing a transferable procedure. Conversely, a learner who pauses during scaffolded practice may be integrating unfamiliar constraints that subsequently produce more reliable understanding. The same observed classroom behavior can therefore support different causal accounts. A rigorous evaluation separates assisted task completion, unassisted near transfer, delayed retention, and more demanding transfer into unfamiliar structures. Studies of faded worked examples and scientific explanation distinguish these endpoints and demonstrate why conclusions based on immediate practice accuracy alone are incomplete Renkl et al. (2002) McNeill et al. (2006).

Guidance is also distributed across time. A teacher may lead students through a model in one lesson, invite partial completion in the next, and organize an independent explanation or solution later. A curriculum or tutoring interface may supply analogous supports across tasks, though the mechanism for deciding when to withdraw them differs. The present research object integrates the decisions about how support is provided, how its suitability is diagnosed, and when learners assume responsibility. Its boundaries are the instructional organization of assistance and its withdrawal; the general architecture of working memory, mastery thresholds, and the entirety of formative assessment remain neighboring scientific objects. The distinction protects the specificity of evidence when different intervention packages are compared van de Pol et al. (2010) Atkinson et al. (2000).

Scaffolding as Contingent and Temporary Support

Scaffolding is a temporally organized relation between the demands of a task, the learner’s current competence, and assistance calibrated to make productive participation possible. In an influential synthesis of teacher–student interaction research, contingency, fading, and transfer of responsibility are identified as defining characteristics van de Pol et al. (2010). Contingency means that support is adapted to observable difficulty and progress. Fading means that the amount, specificity, or availability of help changes over time as capability develops. Transfer of responsibility is the educational end state in which learners increasingly determine relevant steps, monitor their work, and correct errors. The three characteristics are linked through an instructional process rather than represented by an isolated prompt, hint, or attractive visual aid.

The specificity of the term matters because the literature often labels any support as scaffolding. A permanent worksheet can be useful and still offer no evidence of contingency or fading; an instructional video can demonstrate a procedure without adjusting to what any learner currently understands. Puntambekar and Hubscher warned that the expansion of scaffolding from tutoring interactions into software, curricula, and materials often obscured diagnosis and calibrated withdrawal Puntambekar & Hubscher (2005). A scientific account must ask who or what is diagnosing competence, which observations count as sufficient, what support would be provided to a learner at a different level, and how the support will change in response. These questions turn an appealing metaphor into testable constructs.

Contingency cannot be reduced to providing less assistance on every successive attempt. A learner may temporarily need more detailed explanation when confronting a new representation or when earlier apparent fluency proves fragile. Reintroducing assistance in response to demonstrated need is consistent with responsive scaffolding. Equally, the gradual removal of one support may be accompanied by the introduction of another that makes previously hidden disciplinary criteria visible. The evidence on classroom contingency shows that effects depend partly on independent working time and task effort, implying that quantity, timing, and instructional specificity have to be modeled jointly van de Pol et al. (2015).

Sociocultural Foundations and the Development of Assisted Performance

Historical understanding begins with the developmental problem of how social relations become resources for individual action. Vygotsky’s account of the relation between learning and development emphasized abilities evident during cooperation with more capable others and considered instruction in relation to potential development Vygotsky (1978). The zone of proximal development supplies a conceptual orientation: assisted performance can reveal emerging capabilities that independent performance does not yet capture. This idea does not prescribe a particular teaching script or numerical schedule for withdrawing help. It directs attention to the relation between social mediation, task difficulty, and developing competence.

Within this historical tradition, language, signs, demonstration, and collaboration help organize activities that later may be carried out with greater autonomy. A prompt may initially function as an external reminder about the order of a procedure. Over practice, the learner may begin to ask the relevant question without being prompted, use an internal rule, or select an external reference independently. Different forms of independence should be distinguished: unassisted execution is one outcome, the ability to request help strategically is another, and reflective command of the underlying reasons is a further accomplishment. Treating every decrease in overt help seeking as progress would fail to capture these distinctions.

The sociocultural account provides a rationale for studying interactive guidance but is not itself an experimental estimate of its effects. Contemporary teacher–student interaction research makes the distinction operational by coding support in relation to preceding student understanding and by examining subsequent changes in participation and achievement van de Pol et al. (2010). A coherent research history therefore connects theoretical questions about assisted development with later observation, intervention, and transfer designs. The relation is conceptually productive even though the measures, tasks, and populations of contemporary studies differ substantially from the early historical texts.

The Original Tutoring Research and Its Six Functions

The modern scaffolding metaphor became prominent in the 1976 study by Wood, Bruner, and Ross, published as The Role of Tutoring in Problem Solving Wood et al. (1976). Its importance lies in specifying tutor activities during a child’s attempted solution of a task, rather than presenting instruction as the delivery of a fixed amount of information. The authors identified six tutorial functions: recruitment into the task, reduction in degrees of freedom, maintenance of direction, marking critical features, control of frustration, and demonstration. These functions describe different ways of making a challenge approachable while sustaining the learner’s role in achieving the goal.

Recruitment and direction maintenance concern participation and the preservation of the task goal. Reducing degrees of freedom alters the space of decisions the novice must manage. Marking critical features directs attention to differences between an attempted move and the conditions of success. Demonstration makes relevant relations perceptible through a model; frustration control protects the continuation of effort when difficulty becomes discouraging. The functions thus address motivational, cognitive, and regulatory aspects of participation. Their later application to classroom teaching and digital tools has been influential, but the original observations should not be treated as proof that every function improves all outcomes in every domain Wood et al. (1976) van de Pol et al. (2010).

The historical trajectory also requires separating an evocative metaphor from a mechanism. Instructional supports may work by reducing search demands, making criteria visible, coordinating attention, enhancing self-explanation, preventing errors, or preserving willingness to persist. Two supports with the same classroom label can act through different processes. Conversely, a well-chosen question and a partially completed example may both help a learner notice an otherwise neglected principle. This diversity motivates mechanistic comparison and the measurement of learners’ subsequent independent decisions rather than treating the presence of a scaffold as a unitary experimental treatment Reiser (2004).

The Research Map and the Boundaries of Adjacent Constructs

This monograph asks how the provision and adjustment of guidance influence acquisition, understanding, transfer, and independent control. The first research family studies interactional contingency in tutoring and classrooms. A second analyzes worked examples, partial examples, and transitions to problem solving. A third examines supports embedded in inquiry and complex tasks. A fourth evaluates computer-based scaffolding, intelligent tutoring, and the quality of learner adaptation. A final family investigates withdrawal schedules, the distribution of instructional responsibility, and the evidentiary standards needed to compare them. These families share the problem of coordinating assistance and growing competence, although they use different operational definitions van de Pol et al. (2010) Belland et al. (2017).

Cognitive load is a major explanation for why guidance benefits novices, but load is a psychological construct rather than the full object of this dossier. Mastery learning concerns decisions about progression through curriculum and corrective opportunities, whereas fading concerns the composition and dosage of help within and across tasks. Formative feedback provides information used for adjustment, but a feedback statement alone does not establish the ongoing diagnosis, responsiveness, and transfer of task control associated with scaffolding. Collaborative learning introduces peer interactions and group effects that require their own measures. These distinctions permit substantive scientific cross-linking without duplicating neighboring monographs.

Methods in this field include microgenetic study of tutorial exchanges, observational coding of help and response, randomized or quasi-experimental comparisons of instructional conditions, controlled sequences of examples and practice problems, multi-level classroom models, process traces from learning software, and quantitative evidence synthesis. Each method answers a different question. A randomized design can identify an effect of assigning a support package, while interaction coding helps characterize what teachers actually did; delayed transfer tests assess whether observed gains endure beyond the support. Strong synthesis draws these strands together, identifies where theory outruns measurement, and isolates unanswered causal questions McNeill et al. (2006) van de Pol et al. (2015) Lazonder & Harmsen (2016).

The Cognitive Architecture of Early Guidance

Learning a complex procedure imposes demands on attention, working memory, knowledge retrieval, and the coordination of subgoals. Novices often have to inspect many elements separately because they have not yet organized those elements into fluent knowledge structures. They may spend substantial time searching for possible moves without acquiring the relations that make those moves appropriate. In a series of algebra experiments, Sweller and Cooper compared worked examples with conventional problem solving and found advantages for example-based study on structurally similar subsequent problems, while also identifying limits to generalization across forms Sweller & Cooper (1985). The instructional implication is a claim about conditions of learning: when the relevant principles are unfamiliar, externally supplied steps can free cognitive resources for noticing them.

The theoretical account is commonly framed through cognitive load. Guidance can reduce unproductive search and make the structure of a solution available to inspect, while poorly organized guidance can impose additional demands through redundant instructions, fragmented representations, or frequent interruptions. A worked example is consequently more than a shortcut to an answer. It presents a path through a problem whose operations can be compared with underlying rules. The worked-examples review by Atkinson and colleagues explains why example integration, explicit signaling of structure, varying surface details, and placement alongside practice problems are design-relevant Atkinson et al. (2000). None of these devices guarantees that a learner actively interprets the solution.

As expertise grows, the same detailed guidance may cease to help. Kalyuga, Ayres, Chandler, and Sweller describe the expertise reversal effect, in which instructional conditions advantageous for novices can become redundant or counterproductive for more knowledgeable learners Kalyuga et al. (2003). The relevant variable is task-specific prior knowledge, not a global label such as gifted or average. An experienced reader of algebra may still be a novice when interpreting experimental evidence, and a fluent procedure may collapse under a new representation. Consequently, instructional guidance should be adjusted at the level of the knowledge and task being learned; the time spent in a course is only a crude proxy for readiness.

Worked Examples as Explanatory Supports

A worked example displays a resolved problem and the relations between task conditions, intermediate operations, and the final answer. Its instructional value comes from the opportunity to study a meaningful solution without conducting a search for every intermediate move. Examples are particularly useful when learners are unfamiliar with the domain’s permissible transformations, but their effectiveness depends on how the representation draws attention to structurally important features. Atkinson and colleagues synthesized studies of example formats and recommended design features that connect examples with matched practice and conceptual explanations Atkinson et al. (2000). The research supports investigating example structure as an active design variable rather than classifying all shown solutions as equivalent.

An important mechanism is self-explanation. Chi and colleagues analyzed think-aloud protocols of students studying worked mechanics problems and found that more successful learners generated explanations that connected solution steps to principles and more accurately monitored their understanding Chi et al. (1989). This work provides process evidence for an interpretation of example-based learning: the learner must establish why the displayed steps are valid and when they apply. It does not mean that every learner naturally self-explains or that a teacher can infer conceptual understanding from silent attention to a page. Prompting learners to justify a step and to compare two examples is a way of making understanding observable, though any benefit must be measured against added demands.

The distinction between model availability and model dependence is central. An example can support understanding of one problem structure while leaving learners unable to choose among strategies when surface details change. Instruction can therefore sequence comparison, explanation, and increasingly independent selection. Tests need to distinguish copied steps, reconstruction of a demonstrated solution, selection of an appropriate procedure for a new task, and conceptual explanation of why that procedure works. Sweller and Cooper reported strong relevance of structural similarity in their early algebra findings, which cautions against calling an increase in speed on matched problems far transfer Sweller & Cooper (1985).

Procedural Fading and the Transition to Independent Problem Solving

Fading is the planned or responsive withdrawal of instructional assistance as learners acquire the capacity to carry out parts of the task themselves. In the worked-example tradition, a fully solved problem can be followed by a partially completed solution, then by tasks in which progressively more steps must be generated, and finally by independent problem solving. Renkl and colleagues compared faded examples with conventional example–problem pairs across a field experiment and controlled laboratory studies. Their findings favored fading for near-transfer outcomes and indicated that the direction of step removal mattered, with backward fading performing better than forward fading in the studied conditions Renkl et al. (2002).

Backward fading begins by removing later solution steps, so the learner completes the remaining end of a procedure while seeing the preceding subgoals. Forward fading removes earlier steps, requiring learners to initiate a solution before all later dependencies are clear. These are experimentally distinguishable progressions with different implications for search and error. The superiority of one arrangement in a particular domain should not be generalized mechanically to writing, inquiry, or collaborative projects. What matters scientifically is the relationship between the dependencies of the task, the learner's current knowledge, and the demands imposed by the missing steps Renkl et al. (2002) Renkl & Atkinson (2003).

The cognitive-load account of this transition proposes that as schemas develop, solving parts of problems becomes educationally productive rather than merely overwhelming. Renkl and Atkinson argued for a graduated transition in which decreasing instructional completion and increasing learner generation are coordinated with expertise Renkl & Atkinson (2003). A defensible fading schedule therefore requires evidence of adequate understanding, realistic opportunities to practice without prompts, and testing of independence afterward. Assistance may be reintroduced if performance indicates that an assumed prerequisite was never securely acquired. The process is adaptive when withdrawal is justified by current evidence rather than determined only by a calendar or the number of times a worksheet has been assigned.

Contingency, Diagnosis, and the Assistance Decision

Contingent assistance is a decision based on the learner's response, not merely on a teacher's intention to be helpful. The educator must identify the point at which the learner is stalled, decide whether the difficulty reflects a missing fact, an unrecognized relation, a poorly chosen strategy, or uncertainty about the task goal, and select a support suited to that diagnosis. A hint that highlights a critical feature may be sufficient when the procedure is known; a modeled step may be appropriate when its form is unfamiliar; a question about the reason for a choice may uncover a misconception. Studies of interactional scaffolding classify the means and intentions of support to make such distinctions observable van de Pol et al. (2010).

Instructional diagnosis also requires time for students to attempt meaningful action. If a teacher intervenes after every hesitation, frequent assistance can obscure the learner's capacity to plan independently. If help is withheld after repeated unproductive attempts, the difficulty may simply consume time without generating useful information. The experimental classroom study by van de Pol and colleagues found an interaction between quality of contingency and duration of independent group working time, rather than an unequivocal advantage for high-contingency support in every condition van de Pol et al. (2015). The result demonstrates that the effect of assistance cannot be characterized without the temporal context in which it arrives.

Good diagnosis can include a low-cost probe: request a prediction, ask for the next step, examine an intermediate representation, or invite an explanation of why an answer is plausible. The probe should inform the next instructional action. A formative check becomes part of scaffolding when its information genuinely changes the help offered and the responsibility later assigned. This design creates a research challenge because diagnostic accuracy, instructional response, and the learner's change are interdependent. Comparing two teachers who provide different numbers of hints without measuring the difficulties faced by their students confounds adaptation with baseline need.

The Regulation of Attention, Affect, and Agency

Scaffolding can regulate more than solution procedures. The original six functions include securing participation, maintaining direction, and preventing frustration from derailing the activity Wood et al. (1976). A novice may understand an explanation yet fail to persist long enough to apply it, or may devote attention to distracting surface features because the goal has become obscure. A teacher can make the next objective salient, narrow the immediate decision space, and model a constructive response to an error. These are potential mechanisms, whose effectiveness depends on the learner's motivation, task, relationships, and available knowledge.

The motivational value of guidance cannot be inferred from achievement alone. Students may complete more tasks when adults repeatedly direct their attention while becoming less likely to initiate work independently. Conversely, students may value assistance that gives them genuine choice about methods and opportunities to monitor progress. The classroom findings of van de Pol and colleagues separated achievement, task effort, and appreciation of help; more contingent support related to higher appreciation, while effects on achievement varied with working time van de Pol et al. (2015). Evaluation should preserve these distinct outcomes instead of translating positive student reactions directly into learning effects.

Transfer of responsibility includes decisions about how to monitor success, when to seek resources, and how to correct an error. A learner who can decide when expert advice is needed may possess greater agency than one who refuses help to appear independent. The instructional target is therefore competent participation in a practice, including appropriate use of resources and justified judgments. As Reiser argued, scaffolds may deliberately make disciplinary complexity more visible rather than removing all challenge; a task can become locally harder while supporting deeper engagement with the subject Reiser (2004).

Experimental Comparisons of Worked Examples and Problems

The strongest early experimental rationale for guided examples comes from comparisons between learning by inspecting a solution and learning by generating solutions to conventional problems. In algebra instruction, Sweller and Cooper investigated the use of worked examples as alternatives to practice dominated by unsupervised problem solving. Their results established an influential pattern: extensive search for an answer can impose demands that interfere with the acquisition of a useful procedure, whereas studying relevant solution steps can make relations among operations more accessible Sweller & Cooper (1985). Because the experiments were conducted with particular task structures and outcomes, their findings are evidence for conditional instructional mechanisms rather than a universal ranking of teaching methods.

A later research program examined transitions from provided examples to progressively incomplete solutions. Renkl, Atkinson, Maier, and Staley investigated faded examples in experimental and field contexts, including comparisons of solution-step removal sequences and learning outcomes Renkl et al. (2002). The key distinction is between studying a complete model, completing an explicitly selected portion, and generating an entire solution. When tasks depend on a sequence of interrelated transformations, these conditions differ in their demands for search, explanation, and planning. A successful transition requires an opportunity to exercise new control over the procedure without losing access to the principles the previous examples made visible.

These studies have shaped instructional design, but effect interpretation requires attention to the dependent variables. A condition may improve performance on problems closely resembling instruction and show smaller or uncertain effects on new representations, long-delay retention, or independent strategic choice. Research on worked examples should identify the domain, prior-knowledge distribution, total study time, sequencing of examples and problems, and type of outcome assessment. The broader review by Atkinson and colleagues identifies instructional design choices as substantial moderators, so any causal claim about example use alone must specify which forms and comparisons are under examination Atkinson et al. (2000).

Guided Discovery and Inquiry in Quantitative Evidence

The relation between guidance and discovery has generated a long-running debate because the phrase discovery learning encompasses distinct intervention designs. Alfieri and colleagues synthesized experimental evidence concerning unassisted discovery and enhanced discovery formats that supplied forms of feedback, scaffolding, or other support Alfieri et al. (2011). Their findings favored explicit instruction over unassisted discovery in the relevant comparisons while showing benefits for enhanced discovery relative to other conditions. The important scientific distinction is between a learner left to infer all critical relations and a learner actively engaged in inquiry while receiving carefully designed assistance. The labels discovery and instruction do not, by themselves, disclose the instructional mechanisms involved.

Lazonder and Harmsen conducted a meta-analysis focused on guidance within inquiry-based learning, assembling studies that examined the effects of assistance on activity, performance, and learning outcomes Lazonder & Harmsen (2016). This evidence supports the conclusion that inquiry can be strengthened by guidance. Its practical interpretation remains conditional on what guidance provides, whether learners have prerequisite knowledge, and which tests are used. Guiding questions, process constraints, and explicit explanations may serve different purposes; an average across them is useful for identifying a pattern but insufficient for choosing an exact scaffold or removal schedule.

Philosophical and pedagogical disagreements have sometimes been presented as a binary choice between direct teaching and inquiry. Kirschner, Sweller, and Clark argued that minimally guided approaches impose avoidable difficulties on novice learners and defended guidance through cognitive accounts of instruction Kirschner et al. (2006). Hmelo-Silver, Duncan, and Chinn replied that carefully designed problem-based and inquiry learning typically incorporate extensive scaffolding and should be examined as such Hmelo-Silver et al. (2007). These contributions clarify a central methodological demand: an instructional intervention must be described by its actual sequence of support, task demands, and assessment, not classified solely by a broad educational label.

Classroom Contingency and Independent Working Time

Classroom studies bring scaffolding research closer to ordinary teaching while introducing additional variation in tasks, social relations, and instructional implementation. The classroom intervention studied by van de Pol and colleagues examined support contingency in relation to independent working time and outcomes including achievement, effort, and appreciation of assistance van de Pol et al. (2015). The findings suggested that more contingent assistance did not produce a uniform achievement advantage across every instructional arrangement. The amount of time students spent working independently mattered to the observed relationships. This is directly relevant to fading because responsibility cannot transfer if meaningful opportunities for self-directed action are absent.

Contingency is a property of the fit between an instructional move and a learner's preceding understanding. It therefore cannot be estimated by counting teacher questions or calculating total minutes of tutoring. A teacher who makes a complex explanation available precisely when it is needed can provide high contingency even though the intervention is detailed. A teacher who repeats a generic hint with no diagnostic information can provide low contingency despite the appearance of restraint. Research syntheses of teacher–student interaction recommend examining diagnostic activity, the adjustable nature of the help, and changes in the learner's contribution van de Pol et al. (2010).

Generalization from one classroom study should be cautious because student grouping, teacher experience, school context, and the difficulty of independent work influence whether support is productive. Observational coding can show whether assistance followed signs of need, but correlations between such patterns and achievement may also reflect differences in task challenge or teacher allocation of attention. Experimental assignment improves causal inference about an instructional program but cannot automatically identify which interaction inside the program produced the effect. Combining experimental contrasts with process measures and delayed tests provides a stronger explanation of how contingent support affects learning van de Pol et al. (2015).

Fading Written Scaffolds in Scientific Explanation

Complex disciplinary tasks expose a distinctive problem: a support may enable a student to produce a good-looking artifact without developing the conceptual control needed to create one independently. McNeill, Lizotte, Krajcik, and Marx investigated how instructional materials supporting scientific explanations could be faded during a classroom science curriculum McNeill et al. (2006). Their work is important because explanation quality depends on more than executing a routine. Students must construct a claim, select evidence, and justify the relation between that evidence and a scientific principle. Scaffolds can make these components visible and provide an initial structure for their coordination.

In a fading design, students may first encounter written prompts for each element of an explanation, then receive a reduced set of cues, and eventually be required to construct an explanation with little or no procedural prompting. The design question is whether the learner has acquired a stable way of planning and evaluating explanations rather than merely filling in labels. The study by McNeill and colleagues treated the gradual withdrawal of scaffolds as a central experimental variable, allowing comparison of the learning associated with different support sequences McNeill et al. (2006). This work broadens fading beyond mathematical step completion into the practices of reasoning and written argument.

The results are instructive for research design because the visible product and the assessed competence can diverge. Permanently structured worksheets can support fluent production while students remain dependent on their cues; removing too much structure too soon can obscure the disciplinary goal. A rigorous study therefore assesses explanations produced under common unscaffolded conditions and examines their quality with criteria defined independently of the original prompts. It also tracks whether learners apply reasoning principles to unfamiliar scientific questions. Such outcomes distinguish responsibility for an explanation from mere compliance with the format of an instructional material McNeill et al. (2006) Reiser (2004).

Computer-Based Scaffolding and Quantitative Synthesis

Computer-based scaffolds include procedural guides, prompts to plan or monitor activity, explanatory supports, contextual hints, tools for constructing representations, and feedback tied to learner actions. Belland, Walker, Kim, and Lefler synthesized empirical research on computer-based scaffolding in STEM education Belland et al. (2017). Their meta-analysis reported an overall positive average effect, with substantial variation across studies and instructional settings. The aggregate finding is consequential: appropriately organized digital assistance can improve learning outcomes across diverse implementations. Its heterogeneity also demands an investigation of what each tool actually required learners to do.

An especially important question is whether adaptive withdrawal of digital support is superior to keeping that support available. The Belland synthesis did not establish a general fading advantage across the studied computer-based interventions Belland et al. (2017). This finding is compatible with a meaningful role for fading in some tasks, because the interventions vary in quality, timing, learner control, outcome measures, and implementation. It does, however, constrain any universal policy requiring automatic removal of hints after a fixed number of successful responses. Fading should be treated as an instructional hypothesis that needs to be tested for the particular subject, learners, and architecture.

Meta-analytic averages summarize diverse designs rather than define a single mechanism. In a digital environment, fading may remove visual prompts, reduce feedback specificity, delay automated hints, or transfer control over resource selection to the learner. These are substantively different treatments. The outcome may be assessed immediately inside the same software or through later tasks outside it, and the latter presents a stronger test of independence. Consequently, the practical interpretation of meta-analysis depends on intervention descriptions, risk of bias, assessment comparability, and whether teachers or students are able to supplement fading with other supports Belland et al. (2017) Sharma & Hannafin (2007).

Prior Knowledge, Expertise Reversal, and Heterogeneous Treatment Effects

The optimal amount of instructional guidance depends partly on how knowledge is distributed across learners. Expertise reversal is the experimentally motivated phenomenon in which an instructional feature that reduces difficulty for novices becomes unhelpful or disruptive for learners who can already organize the task effectively Kalyuga et al. (2003). An explanation that resolves an unfamiliar relation may be valuable to one student and repetitive to another. In technical tasks, unnecessary repetition may compete with the learner's self-directed processing. These effects imply that prior knowledge should be measured at the level of the relevant domain and task before comparing forms of support.

Expertise is dynamic and multidimensional. A learner may know the facts involved in a task but remain uncertain about their coordination, or may execute a familiar sequence while lacking the conceptual resources needed for transfer. One assessment can therefore overestimate readiness for independence. The framework developed by Koedinger, Corbett, and Perfetti encourages specifying the component of knowledge to be acquired, the learning process thought to change it, and the instruction likely to produce that change Koedinger et al. (2012). This is a more rigorous basis for adaptation than assuming that school grade, age, or previously obtained test scores fully determine scaffolding needs.

Heterogeneity also limits general statements about who benefits from fading. In an experiment, the effect of guidance for a novice group may differ from that for a more knowledgeable group; in a classroom, learners move between levels as they acquire new knowledge. These two sources of variation need different designs: pretest-based moderator analysis can examine prior competence, while repeated within-person assessments can investigate transitions. A strong hypothesis is that assistance should decline when its informational value declines, but support that reveals hidden criteria may remain useful even after procedures become fluent. The hypothesis is grounded in expertise-sensitive theory and requires direct testing for each class of tasks Kalyuga et al. (2003) Renkl & Atkinson (2003).

Instructional Sequencing in Mathematics and Procedural Domains

Mathematical learning offers a particularly clear setting for investigating instructional guidance because many tasks have well-defined intermediate operations and testable dependencies. A novice working through algebra must recognize symbols, select a transformation, anticipate its consequences, and check whether the transformed expression preserves the required relation. A worked example makes one plausible sequence visible and allows the learner to relate the steps to general rules. Experiments with worked examples established an advantage in appropriately chosen early learning conditions, though the specificity of the training tasks influences how broadly the result can be applied Sweller & Cooper (1985). The instructional choice is consequently between different opportunities to acquire explanatory structures, not simply between passive watching and active working. Rosenshine's research-informed instructional synthesis presents modeling, small steps, guided practice, and checks for understanding as practical components of such structured teaching; it is a practitioner synthesis rather than a newly randomized estimate Rosenshine (2012).

A practical instructional sequence may begin with modeling and teacher explanation, continue with paired comparison of similar but meaningfully different examples, move to partially completed solutions, and culminate in independent choice of methods. Each stage introduces a distinct demand. Comparison tests structural recognition; completion of a missing step tests partial procedural command; solving an entire problem tests integration and planning. Renkl and colleagues investigated the smooth transition from example study to problem solving and reported benefits under the conditions they studied Renkl et al. (2002). The design should be monitored with short probes and delayed assessments rather than imposing the same tempo on every learner.

Procedural fluency is not sufficient evidence of conceptual transfer. Students may memorize which step follows another in a familiar format while failing to identify the circumstances under which a rule applies. Self-explanation can turn a displayed procedure into a claim that the learner can defend; the think-aloud findings of Chi and colleagues suggest that such explanations distinguish more successful study of solved problems Chi et al. (1989). A rigorous mathematics curriculum should therefore include tasks that require explaining, diagnosing invalid steps, choosing among multiple solution routes, and solving new instances without the model.

Scaffolding Inquiry and Disciplinary Explanation

Inquiry learning asks students to generate and evaluate questions, evidence, and explanations. The task space is less predictable than that of a fully specified computational exercise, so scaffolds often organize planning and disciplinary criteria instead of simply displaying an answer. Reiser described two broad mechanisms: structuring work by making complex tasks manageable and problematizing work by making substantive disciplinary issues visible Reiser (2004). These processes can operate together. An organizer may reduce the burden of remembering procedural steps while a question about conflicting evidence increases attention to the underlying scientific problem. Evaluating the tool requires understanding which activity is being made easier and which is being made more intellectually demanding.

An inquiry scaffold might ask students to record competing explanations, justify a measurement choice, or link evidence explicitly to a claim. At the outset, such prompts can make important practices accessible to novices. Later, some prompts may be removed while requirements for evidence quality remain in place. The fading study by McNeill and colleagues examined students' construction of scientific explanations, showing how instructional supports can be varied within an authentic curricular task McNeill et al. (2006). The relevant endpoint is independent argumentation according to disciplinary standards, not a reduction in the total number of text boxes or questions presented on a worksheet.

Research on inquiry needs to distinguish minimal guidance from guided inquiry. Hmelo-Silver, Duncan, and Chinn emphasized that problem-based and inquiry approaches typically contain scaffolding practices that are central to their design Hmelo-Silver et al. (2007). Lazonder and Harmsen's evidence synthesis further demonstrates that guidance deserves direct analysis as an instructional component Lazonder & Harmsen (2016). Strong studies document the actual assistance supplied, including its content, timing, teacher mediation, and accessibility, while testing student understanding outside the particular inquiry format. The broad name of a pedagogy should never substitute for a description of its causal ingredients.

Teacher–Student Dialogue, Group Work, and Classroom Feasibility

Teachers often adjust help within a conversation whose meaning depends on the learner's preceding utterance. A student who proposes a partial explanation reveals a different need from a student who has not yet identified the task goal. In responsive teaching, the next question can focus attention on an overlooked relation, request a justification, or supply the missing representation. The synthesis by van de Pol, Volman, and Beishuizen shows why scaffolding is commonly understood through contingency, fading, and transfer of responsibility van de Pol et al. (2010). These properties are sequential: an observation informs support, and subsequent responses reveal whether the learner is ready for an altered level of assistance.

Classroom conditions complicate the translation of one-to-one tutoring principles. One teacher may interact with several groups, while assistance to one learner affects the time available to others. Students themselves may exchange explanations, and a brief teacher intervention may alter the group's division of labor. The intervention studied by van de Pol and colleagues suggests that independent working time plays a role in the achievement effects of contingent support van de Pol et al. (2015). This creates a practical requirement for classroom design: students must receive meaningful intervals in which they attempt, monitor, and revise work without continual teacher direction.

Research should specify the unit of analysis. An individual learner's gain cannot always be attributed to an individual hint when classmates contribute to the solution or classroom norms organize participation. Cluster-randomized trials can reduce some threats to causal identification, while detailed coding can separate teacher assistance from peer contributions. A complete account also examines who receives advanced questioning, who gets simplified steps, and whether apparently differentiated support distributes opportunities for high-level reasoning equitably. These are design and measurement questions for the empirical study of instructional systems, rather than facts that can be inferred from scaffolding terminology alone.

Digital Prompts, Adaptive Interfaces, and Learning Analytics

Technology-enhanced scaffolding can preserve supports across tasks, offer prompts at specified decision points, and record how learners respond. Sharma and Hannafin reviewed the role of scaffolding in technology-enhanced learning environments and distinguished the resources through which learners receive assistance Sharma & Hannafin (2007). A digital prompt may be fixed, learner-requested, triggered by behavior, or selected by an adaptive model. These forms differ in their assumptions about learner control and about the validity of the information used to diagnose need. Simply presenting hints on a screen is therefore insufficient evidence that a tool provides individualized scaffolding.

Some digital systems adapt based on error frequency, latency, previously mastered steps, or patterns of help seeking. These signals are convenient but ambiguous. Slow responding can reflect thoughtful evaluation rather than confusion; rapid completion can reflect guessing or copying. A system that removes help after several correct responses may mistake short-term familiarity for durable knowledge. Adaptive design needs validated measures of the construct it claims to estimate and assessments of performance when assistance is absent. The knowledge-learning-instruction framework identifies why the target knowledge component must be explicit before a model's predictions can guide an instructional move Koedinger et al. (2012).

Computer-based scaffolding research is broad enough to support quantitative synthesis, but the heterogeneity reported by Belland and colleagues cautions against treating every prompt, tool, or adaptive interface as the same intervention Belland et al. (2017). For each system, researchers should report the target domain, learner sample, input signals, support policy, availability of human assistance, opportunities to ignore or request help, and the tests used to establish transfer. Digital systems also make it possible to compare fixed schedules with performance-sensitive fading under controlled conditions. Such comparisons should be judged by learning quality and student autonomy, not only by engagement metrics or time spent in the interface.

Artificial Tutoring and Instruction in the Artificial Era

Artificial tutoring systems extend the question of instructional guidance into adaptive dialogue. A system may generate explanations, ask diagnostic questions, provide partial solutions, and revise its responses after learner feedback. A randomized study reported by Kestin and colleagues evaluated an AI tutor against an in-class active-learning comparison in an authentic educational setting Kestin et al. (2025). The study is relevant to instructional design because it investigates a concrete implemented tutoring condition and its measured outcomes. It does not settle whether every conversational system, subject, or learner population benefits from the same architecture. The specificity of the design must travel with any interpretation of its result.

Generative tutoring introduces a difficult measurement problem: a convincing conversational explanation can coexist with an incorrect inference about what the learner knows. A model may supply a polished step that advances the immediate task while leaving the student's later independent solution unimproved. Effective evaluation should therefore include common tests taken without the tutor, delayed retention, accurate diagnosis of misconceptions, and explicit records of when support was removed or reintroduced. System-generated confidence should not be substituted for observed reliability. The broader knowledge-learning-instruction perspective helps formulate experiments that compare policies for asking, explaining, modeling, and withholding help Koedinger et al. (2012).

The Artificial Era supplies a wider research horizon for this subject because instructional responsibility may be shared across students, human teachers, digital tools, and artificial agents. This changes the organization of assistance and the allocation of decisions about relevance, timing, and evaluation. It is a research program, not an empirical conclusion that a particular artificial tutor possesses the learner's psychological understanding. Useful future tests compare human and artificial scaffolding policies, measure learners' ability to contest or correct advice, and examine whether the system leaves learners more able to reason independently. The scientific unit is the instructional interaction and its verified outcomes.

Development, Accessibility, and Contextual Adaptation

Learners enter instructional situations with different language resources, sensory access, relevant prior knowledge, and experiences of schooling. These differences influence which aspects of a task are needlessly difficult and which constitute the substantive learning goal. Vygotsky's work offers a conceptual basis for treating assistance as situated within social and developmental relations Vygotsky (1978). It does not authorize predicting support needs solely from chronological age or a fixed developmental category. A learner can be highly competent in one practice and need extensive mediation when task language or representation changes.

Accessibility requires distinguishing the supports that enable equitable access from the supports whose temporary removal is being studied. Captions, alternative text, screen-reader compatibility, language access, and accommodations may remain essential even after conceptual mastery has developed. Removing them would change access to information rather than test independent competence in the targeted domain. The general framework of scaffolding is therefore strongest when the instructional goal, interface, learner's mode of participation, and legitimate accommodations are specified before fading decisions are made Puntambekar & Hubscher (2005) Sharma & Hannafin (2007).

Generalization across countries, languages, and institutions presents an additional evidence challenge. A practice developed for one classroom structure may rely on teacher time, familiar modes of dialogue, or assessment conventions unavailable elsewhere. Learners' willingness to request help may also reflect classroom norms rather than a direct indicator of understanding. Comparative research should measure the same targeted capability in culturally and linguistically appropriate ways and document the instructional resources available in each setting. Without such evidence, universal claims about the speed or desirability of withdrawal exceed the research base.

Operational Definitions and Measurement of Scaffolding

A defensible operational definition begins by separating support provision, support quality, assistance withdrawal, and learner independence. Support provision can be recorded as the type, timing, and amount of explanation, modeling, prompts, or feedback made available. Support quality concerns its fit to a diagnosed learning problem. Withdrawal concerns a change in accessibility, specificity, frequency, or control of help. Independence concerns performance in conditions where the previously available support is absent or learner-directed. These components are related but empirically separable, and the synthesis of teacher–student interaction by van de Pol and colleagues provides a framework for their measurement van de Pol et al. (2010).

Observational measures may code the learner's difficulty, the teacher's response, and the subsequent learner move. Experiments may randomize access to a support or its fading schedule. Learning software may record hint requests, task steps, and revisions, but the logs have meaning only in relation to valid task criteria. Outcome measures should distinguish immediate assisted production from unassisted reconstruction, delayed retention, and transfer. A scaffold that yields excellent completed work while remaining permanently present establishes a different result from one that helps learners later reason independently. Investigations of scientific explanations show the importance of assessing what remains after written prompts are withdrawn McNeill et al. (2006).

Construct validity depends on clear counterfactuals. An observed decrease in help seeking could signify growing competence, withdrawal of available resources, embarrassment, disengagement, or a change in the task. A rise in student completion rates could follow improved understanding or a narrower space of choices. The study of classroom contingency and independent working time demonstrates that even a sophisticated instructional variable operates within a larger sequence of activity van de Pol et al. (2015). Researchers should predefine the outcomes their intervention is intended to change, report implementation fidelity, and examine alternative explanations before attributing improvement to fading itself.

Causal Identification and Limits of the Existing Evidence

Randomized assignment supports causal inference about the compared instructional packages when implementation, attrition, and outcome assessment are adequately controlled. Yet a package may differ from its comparison in teacher attention, instructional time, information content, feedback, or opportunities for independent practice. If an experiment compares a carefully guided program with an under-resourced alternative, it identifies an effect of the package and cannot independently identify the contribution of a particular hint or fading operation. Factorial designs and carefully matched controls are needed to separate the effects of support content, dosage, and withdrawal policy. The wide range of instructional formats examined in research syntheses makes this distinction essential Alfieri et al. (2011) Belland et al. (2017).

Several evidence limitations recur. Research often concentrates on particular disciplinary tasks, short instructional periods, or accessible student groups. Outcomes are sometimes measured near the end of instruction, when the apparent advantage may be sensitive to the match between teaching format and test items. Classroom trials also face teacher effects, social interference, and differences in students' opportunities to participate. In digital settings, implementation may be unusually standardized while the relation to everyday classroom conditions becomes less direct. The broad positive averages in scaffolding meta-analyses consequently support a family of interventions while leaving specific policy choices open Lazonder & Harmsen (2016) Belland et al. (2017).

Reporting standards should include the study population, prior knowledge, task complexity, duration, assignment unit, instructor training, actual support delivered, adherence to fading rules, missing data, and the sensitivity of the outcomes to assistance. Researchers should distinguish preregistered hypotheses from exploratory moderator analysis and avoid interpreting observational associations as causal estimates. Independent replication and studies with delayed common tests are particularly valuable. A research synthesis can identify convergent patterns without presenting itself as a protocolized systematic review; the present monograph is an evidence-informed critical synthesis rather than a new quantitative meta-analysis.

The Central Fading Problem and Its Competing Solutions

The central unresolved question is how an instructional system should decide that a support has done enough work and can be withdrawn. A fixed schedule offers consistency and ease of implementation, but assumes that learners acquire the relevant competence at roughly the same rate. Performance-sensitive fading can respond to demonstrated progress, but risks inferring mastery from narrow tasks or noisy behavioral signals. Learner-controlled fading respects strategic self-regulation, yet the learner may lack accurate information about the limits of current understanding. Each solution encodes a hypothesis about how assistance, information, and agency should be coordinated.

Research on faded examples provides evidence that structured transitions from examples to problems can be useful, including effects of the order in which solution steps are removed Renkl et al. (2002) Renkl & Atkinson (2003). Classroom science research addresses the removal of prompts for disciplinary explanation rather than ordered mathematical steps McNeill et al. (2006). Meanwhile the computer-based meta-analysis by Belland and colleagues does not establish a universal advantage for fading over persistent scaffolds Belland et al. (2017). The findings can be integrated when fading is treated as a set of context-specific interventions rather than one mechanism with an invariant effect.

A stronger decision rule would combine evidence of understanding, tests of unaided performance, and the informational value of additional help. It could make support optional rather than simply eliminate it; it could fade one feature while maintaining access accommodations or reference resources; and it could restore an explanation when learners encounter a new kind of problem. These are testable design proposals, not demonstrated universal prescriptions. Experiments should compare them with transparent fixed schedules and with persistent support, while measuring both sustained competence and the learner's capacity to select help appropriately.

Retention, Transfer, and the Meaning of Independent Competence

Learning is inferred more strongly when a learner can perform after a meaningful delay and under conditions different from those used during instruction. Near transfer may require applying familiar procedures to new examples, whereas broader transfer requires identifying principles across altered surfaces, goals, or representations. In both cases the critical question is what remains available when the training scaffold disappears. The research on worked examples, fading, and scientific explanation shows why immediate task success cannot stand alone as proof of flexible competence Sweller & Cooper (1985) Renkl et al. (2002) McNeill et al. (2006).

Independent competence also includes judgment about limits. In authentic educational and professional environments, people use books, collaborators, assistive technologies, and digital systems as resources. The capacity to choose a relevant resource, evaluate it, and detect an incorrect suggestion may be more educationally meaningful than performing every action without external materials. Responsibility can thus shift from following prescribed prompts to selecting and managing the conditions of one's own inquiry. The distinction is especially important when a computer-based scaffold is built into the environment learners will continue using after formal instruction Sharma & Hannafin (2007) Belland et al. (2017).

Research should measure both autonomy and effective resource use. A repeated assignment can compare performance with standard resources available to all participants, while a second assessment removes training-specific prompts to test their contribution. Process measures may capture help-seeking decisions, error diagnosis, explanatory quality, and corrections after feedback. These indicators should be interpreted in light of the domain and instructional purpose. A learner who uses a trustworthy external reference at the right moment may display a high level of independent regulation; a learner who completes a memorized sequence without understanding may display a more limited form of independence.

An Integrated Research Agenda for Instructional Assistance

The literature supports a research agenda organized around mechanisms and decisions. One program should identify what learners acquire from different kinds of examples and explanations, using measures of self-explanation and conceptual change rather than only final answers Chi et al. (1989) Atkinson et al. (2000). Another should compare guidance policies matched to task-specific knowledge, documenting when expertise reversal occurs and whether diagnostic assessments predict the point at which detailed help becomes redundant Kalyuga et al. (2003). A third should measure how teacher contingency influences independent working time, persistence, and delayed achievement under realistic classroom constraints van de Pol et al. (2015).

Digital tutoring makes it feasible to compare explicit fading rules, learner-controlled hint access, and policies that reinstate assistance after an informative error. Such experiments should report how a system infers knowledge, distinguish predictive accuracy from educational benefit, and include common unassisted tests. Large-scale synthesis needs more transparent reporting of intervention ingredients, moderation by domain and prior knowledge, and follow-up beyond the immediate teaching environment Belland et al. (2017). Research on artificial tutoring should similarly evaluate whether fluent assistance transfers responsibility to the student, rather than measuring conversational smoothness as an educational endpoint Kestin et al. (2025).

At an interdisciplinary level, scaffolding connects developmental theories of mediated activity, cognitive science of procedural learning, educational psychology of classroom interaction, and human–artificial research on adaptive assistance. These connections yield testable questions: which support components are functionally interchangeable across human and artificial tutors, what forms of diagnosis justify withdrawal, how does learner control interact with accuracy, and how durable is competence after a change of instructional partner? The scientific synthesis favors specifying interactions, tasks, and outcomes at the level where these questions can be answered.

Conclusion

Guided instruction, scaffolding, and fading form a coherent research object concerned with the changing relationship between task demands, learner competence, and instructional assistance. Historical accounts of assisted development and tutoring supplied enduring concepts; experimental studies of worked examples and procedural fading made aspects of the learning process measurable; classroom research operationalized contingency and independent working time; and research on inquiry and computer-based support broadened the range of tasks and instructional environments Vygotsky (1978) Wood et al. (1976) Renkl et al. (2002) van de Pol et al. (2010) Belland et al. (2017). The resulting field has substantive empirical foundations and a clear set of unresolved causal questions.

Across this literature, structured and responsive guidance can improve learning, particularly when novices face tasks whose critical relations are difficult to infer unaided. Successful instructional design also requires opportunities for students to explain, practice, correct, and test their own understanding. Evidence for the value of support does not imply that one fading schedule is best for every learner or discipline. The appropriate sequence depends on what the task requires, which knowledge has been acquired, what form of assistance is still informative, and how independence will be evaluated Alfieri et al. (2011) Lazonder & Harmsen (2016) Kalyuga et al. (2003) Belland et al. (2017).

The central scientific principle is to align assistance with developing competence while treating transfer of responsibility as an outcome that must be demonstrated. The next stage of research should establish more precise, replicable relationships among diagnosis, support policy, learner control, and retained capability. This applies to ordinary classrooms, inquiry environments, adaptive digital systems, and emerging artificial tutoring relationships. The strongest instructional systems will be those whose effects are verified through learners' increasingly justified, flexible, and enduring command of the knowledge they are learning.

References

  • Alfieri, L., Brooks, P. J., Aldrich, N. J., & Tenenbaum, H. R. (2011). Does discovery-based instruction enhance learning? Journal of Educational Psychology, 103(1), 1–18. https://doi.org/10.1037/a0021017
  • Atkinson, R. K., Derry, S. J., Renkl, A., & Wortham, D. (2000). Learning from examples: Instructional principles from the worked examples research. Review of Educational Research, 70(2), 181–214. https://doi.org/10.3102/00346543070002181
  • Belland, B. R., Walker, A. E., Kim, N. J., & Lefler, M. (2017). Synthesizing results from empirical research on computer-based scaffolding in STEM education: A meta-analysis. Review of Educational Research, 87(2), 309–344. https://doi.org/10.3102/0034654316670999
  • Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Self-explanations: How students study and use examples in learning to solve problems. Cognitive Science, 13(2), 145–182. https://doi.org/10.1207/s15516709cog1302_1
  • Hmelo-Silver, C. E., Duncan, R. G., & Chinn, C. A. (2007). Scaffolding and achievement in problem-based and inquiry learning: A response to Kirschner, Sweller, and Clark (2006). Educational Psychologist, 42(2), 99–107. https://doi.org/10.1080/00461520701263368
  • Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38(1), 23–31. https://doi.org/10.1207/S15326985EP3801_4
  • Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning: An RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, Article 17458. https://doi.org/10.1038/s41598-025-97652-6
  • Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problem-based, experiential, and inquiry-based teaching. Educational Psychologist, 41(2), 75–86. https://doi.org/10.1207/s15326985ep4102_1
  • Koedinger, K. R., Corbett, A. T., & Perfetti, C. (2012). The knowledge-learning-instruction framework: Bridging the science-practice chasm to enhance robust student learning. Cognitive Science, 36(5), 757–798. https://doi.org/10.1111/j.1551-6709.2012.01245.x
  • Lazonder, A. W., & Harmsen, R. (2016). Meta-analysis of inquiry-based learning: Effects of guidance. Review of Educational Research, 86(3), 681–718. https://doi.org/10.3102/0034654315627366
  • McNeill, K. L., Lizotte, D. J., Krajcik, J., & Marx, R. W. (2006). Supporting students' construction of scientific explanations by fading scaffolds in instructional materials. Journal of the Learning Sciences, 15(2), 153–191. https://doi.org/10.1207/s15327809jls1502_1
  • Puntambekar, S., & Hubscher, R. (2005). Tools for scaffolding students in a complex learning environment: What have we gained and what have we missed? Educational Psychologist, 40(1), 1–12. https://doi.org/10.1207/s15326985ep4001_1
  • Reiser, B. J. (2004). Scaffolding complex learning: The mechanisms of structuring and problematizing student work. Journal of the Learning Sciences, 13(3), 273–304. https://doi.org/10.1207/s15327809jls1303_2
  • Renkl, A., Atkinson, R. K., Maier, U. H., & Staley, R. (2002). From example study to problem solving: Smooth transitions help learning. Journal of Experimental Education, 70(4), 293–315. https://doi.org/10.1080/00220970209599510
  • Renkl, A., & Atkinson, R. K. (2003). Structuring the transition from example study to problem solving in cognitive skill acquisition: A cognitive load perspective. Educational Psychologist, 38(1), 15–22. https://doi.org/10.1207/S15326985EP3801_3
  • Rosenshine, B. (2012). Principles of instruction: Research-based strategies that all teachers should know. American Educator, 36(1), 12–19, 39. https://www.aft.org/ae/spring2012/rosenshine
  • Sharma, P., & Hannafin, M. J. (2007). Scaffolding in technology-enhanced learning environments. Interactive Learning Environments, 15(1), 27–46. https://doi.org/10.1080/10494820600996972
  • Sweller, J., & Cooper, G. A. (1985). The use of worked examples as a substitute for problem solving in learning algebra. Cognition and Instruction, 2(1), 59–89. https://doi.org/10.1207/s1532690xci0201_3
  • van de Pol, J., Volman, M., & Beishuizen, J. (2010). Scaffolding in teacher–student interaction: A decade of research. Educational Psychology Review, 22(3), 271–296. https://doi.org/10.1007/s10648-010-9127-6
  • van de Pol, J., Volman, M., Oort, F., & Beishuizen, J. (2015). The effects of scaffolding in the classroom: Support contingency and student independent working time in relation to student achievement, task effort and appreciation of support. Instructional Science, 43(5), 615–641. https://doi.org/10.1007/s11251-015-9351-z
  • Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes (M. Cole, V. John-Steiner, S. Scribner, & E. Souberman, Eds.). Harvard University Press. https://doi.org/10.2307/j.ctvjf9vz4
  • Wood, D., Bruner, J. S., & Ross, G. (1976). The role of tutoring in problem solving. Journal of Child Psychology and Psychiatry, 17(2), 89–100. https://doi.org/10.1111/j.1469-7610.1976.tb00381.x