Prompting learners to self-explain, one generative strategy, improves learning by a moderate average amount; other generative activities are not tested by the evidence recorded here¶

Claim · All claims
Evidence · 1 study · quant-synthesis r? · q4 · i2 medium · n=69 effect sizes (64 research reports)
Currency · newest study 2018 · not yet checked against its sources · not reviewed by a person

Generative processing means learners actively construct connections between new material and their prior knowledge, by summarizing, self-explaining, mapping or generating answers. The one entry recorded, a meta-analysis of 64 reports (Bisra et al. 2018), tests self-explanation prompts only, and finds a moderate average benefit (g = 0.55) over conditions without them; whether summarizing, mapping or generating answers do the same is not shown here. The claim concerns the act of generation itself; comparisons must control for study time, since generative activities typically take longer than passive study.

Subclaims¶

q4 i2 Prompting learners to self-explain — a canonical generative strategy that forces selecting, organizing, and integrating information — produces a moderate learning benefit over passive study or problem-solving without such prompts, across a wide range of instructional conditions. → Bisra et al. 2018

q4 i2 Across 69 effect sizes from 64 research reports meeting inclusion criteria, participants who received self-explanation prompts while studying or solving problems showed better learning outcomes, with a random-effects weighted mean g = .55. → Bisra 2018

q4 i2 Prompting learners to self-explain — a core elaborative-encoding operationalization — produces a medium-sized average improvement in learning outcomes across a wide range of instructional conditions, tasks, and educational levels. → Bisra et al. 2018

Evidence¶

Bisra et al. 2018¶

Bisra, K., Liu, Q., Nesbit, J. C., Salimi, F., & Winne, P. H. (2018). Inducing Self-Explanation: a Meta-Analysis. Educational Psychology Review, 30(3), 703–725. doi:10.1007/s10648-018-9434-x

q4 · meta-analysis (random-effects model) · i2 · medium effect, g=0.55 · n=69 effect sizes (64 research reports) · quant-synthesis · r?

A meta-analysis of studies that induced self-explanation — a generative strategy in which learners produce inferences about causal connections or conceptual relationships in to-be-learned material — while studying text, worked examples, or solving problems, compared with matched conditions without self-explanation prompts. Pooling 69 effect sizes from 64 research reports with a random-effects model, the authors found an overall weighted mean effect of Hedges' g = 0.55 favoring self-explanation. The benefit held across 20 coded moderators (task type, subject area, level of education, type of inducement, treatment duration), leading the authors to describe self-explanation prompts as "a potentially powerful intervention across a range of instructional conditions."

Studies bearing on this claim¶

Does having learners generate their own elaborations improve learning, compared with no elaboration or with elaborations supplied to them? This table gathers the studies behind this claim and the 15 narrower claims that bear on it: 15 distinct studies (5 causal, 4 quant-synthesis, 3 review, 1 theoretical, 1 associational, 1 design), q1–q4, 5 of 15 with an effect size coded. One row per study, codes as each entry records them; no effect is averaged and no verdict is drawn. Built by scripts/synthesize_claim.py from eval/synthesis/; a row that looks wrong is fixed on the claim it comes from.

Discussion¶

Mechanism. Generative processing is the productive engagement hypothesized by cognitive load theory and generative models of comprehension: learners select relevant information, organize it into a coherent structure, and integrate it with prior knowledge. Activities that prompt this work — self-explanation, summarizing, concept mapping, drawing, generating examples — are predicted to outperform passive study of identical content because they force construction rather than recognition. This claim is the umbrella under which several more specific claims on this wiki sit: annotating, activation, and active learning can all be read as particular instantiations of generative processing.

Moderators and boundary conditions. The benefit depends on the learner actually generating, not just being given the opportunity to. Learners with low prior knowledge may lack the schema to generate useful connections, so generation can impose extraneous load rather than germane load — a pattern consistent with the expertise reversal effect. Poorly designed generation tasks (e.g., copying, or generating before any input is understood) can also backfire. Quality of the generated product, time on task, and alignment between the generation task and the criterion test all plausibly moderate the effect; these need to be established by evidence before strong design prescriptions follow. A recurring confound in this literature is time on task: generative activities typically take longer than passive study, so comparisons must control for study time to isolate the effect of generation itself.

Relation to load management. Generation is not free: it consumes working memory resources. It therefore pairs naturally with chunking and other load-management measures — the claim is that generative effort pays off only when the material itself does not overwhelm the learner, as captured by cognitive overload degrades learning. Designers should treat generative tasks as an investment of limited capacity, sequenced after basic comprehension of the input rather than instead of it.

Open questions. Which generation activities are most efficient per unit of time, how generation interacts with worked examples and fading, and how effects scale from lab tasks to classroom curricula all remain to be documented on this page. One entry is recorded, a meta-analysis of self-explanation prompts; for other generative activities, this page is a framing claim whose specific instantiations carry their own evidence (summarizing, for instance, did worse than rereading in one multi-document study: summarization).

Merged from “Generative Learning Improves Retention” (generative-learning-improves-retention): Mechanism. Generative activities are hypothesized to work by forcing learners to construct relations between new material and prior knowledge, rather than reproducing surface text. This aligns with the broader account in Cognitive Load Theory: generation imposes effortful processing that supports schema construction, but only when the extra load is germane rather than extraneous — see Cognitive overload degrades learning.

Moderators and boundary conditions. Generation is not uniformly beneficial. Learners need sufficient prior knowledge to generate accurate products; when they lack it, generative tasks can produce errors or flounder, and providing structure (prompts, sentence starters, worked models) becomes necessary. The activity must also actually require transforming meaning — copying, highlighting, or verbatim note-taking look generative but do not produce the same benefit. Task–learner fit matters: the same prompt that helps one learner may be redundant or overwhelming for another, echoing the expertise-reversal pattern documented in the expertise reversal effect.

Retention versus transfer. The folded page's claim concerned retention. Generative strategies are often expected to support transfer and inference as well, but the meta-analysis recorded above measured self-explanation, one generative strategy, so the strength of that expectation for other strategies is not established here; retention benefits are typically the better-established outcome in this literature.

Open questions. Most of the evidence base compares generation against passive control conditions; fewer studies test which generative activity is best for a given material type, or how benefits persist over delay intervals versus immediate tests. The recorded evidence covers self-explanation only, so confidence in the broader claim about generative strategies should stay moderate.

Merged from “Generative Learning Improves Comprehension” (generative-learning-improves-comprehension): Mechanism. Generative activities are hypothesized to work by forcing learners to select relevant information, organize it into a coherent structure, and relate it to prior knowledge — the three processes in generative models of comprehension (Fiorella & Mayer's selecting–organizing–integrating framework). This aligns with the broader principle that active learning outperforms passive reception [+M], and with activation of prior knowledge as a precondition for meaningful integration [+M].

Not all generation is equal. The benefit depends on the quality of the generative process. Copying text verbatim or underlining involves little transformation and yields little gain [~M], whereas activities that require constructing relations — such as self-explaining, concept mapping, or annotating — impose the selection–organization–integration cycle that drives comprehension [+M]. Fiorella and Mayer distinguish "summarizing" and "mapping" (organizing strategies) from "self-explaining" and "teaching" (integrating strategies); both families outperform passive study, but through different process routes.

Moderators and boundary conditions. Generation is effortful; learners with limited prior knowledge or high working-memory demands may benefit from more scaffolded generative tasks, consistent with cognitive load theory [~M]. Learners also tend to prefer rereading over generating, even though rereading is less effective — so designers should not treat learner preference as a guide [-W]. The benefit is strongest for measures of comprehension and transfer rather than verbatim recall [+M].

Open questions. Which generative strategy is optimal for a given domain, learner profile, and outcome measure remains unsettled; comparative studies often find small differences among well-chosen strategies, suggesting that any activity that reliably triggers the underlying processes captures most of the benefit. Studies still need to be added to the Evidence section before this claim can carry an evidence-strength rating.

Merged from “Elaborative Encoding Improves Retention” (elaborative-encoding-improves-retention): Mechanism. Elaboration is typically explained through the depth-of-processing framework: richer, more meaningful processing creates more retrieval routes than shallow repetition. In practice, elaboration overlaps with several related operations — activation of prior knowledge, analogical reasoning, and chunking — so isolating its unique contribution is difficult. Elaborative interrogation ("why is this true?") and self-explanation are the two most-studied operationalizations; both ask learners to integrate new material with what they already know rather than re-read it.

Boundary conditions. Elaboration is effortful and depends on working-memory capacity; learners with low prior knowledge may generate inaccurate elaborations that are then encoded and later retrieved as errors. Quality of elaboration matters more than quantity — self-generated connections to relevant prior knowledge support retention, while forced or superficial elaboration ("this word reminds me of my dog") may add processing load without adding meaning. This connects to cognitive overload: elaboration demands working-memory resources that overload can consume, so prompting learners to elaborate dense or unfamiliar material can backfire. Designers should therefore scaffold elaboration — provide partial prompts, worked examples of good elaborations, or advance organizers that supply the knowledge to connect to — rather than issuing open-ended "explain this to yourself" instructions to novices.

Moderators to expect. When evidence is populated, likely moderators include prior knowledge (elaboration should benefit learners who have relevant knowledge to connect to), accuracy of self-generated elaborations (errors encoded during elaboration are later retrieved as errors), and outcome type — elaboration plausibly supports retention and comprehension more than it supports transfer, where analogical reasoning across cases does more work. Time pressure is a further moderator: elaborative strategies are slower than restudy, so under fixed study time the benefit can shrink or reverse.

Open questions. The evidence base for this claim still needs to be populated — studies comparing elaborative interrogation and self-explanation against restudy, and work on when elaboration helps retention versus transfer, should be added before this page is treated as authoritative.

Related Claims¶

• Active learning improves exam performance. — active engagement in class produces better outcomes than lecture alone
• Activation improves learning. — activating prior knowledge supports integration of new material
• Annotating improves learning. — annotation is a concrete generative activity during reading
• Chunking reduces working memory load. — managing load is a precondition for productive generative effort
• Cognitive overload degrades learning. — generation helps only when working memory is not overwhelmed
• 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. — related
• Students who generate their own research question are apt to be more invested and more engaged — related
• Students with more controlled interaction patterns in iSTART-2 generated higher-quality self-explanations than students with more random patterns — a narrower finding that bears on this claim
• Learners' beliefs about a medium and its processing demands influence the mental effort they invest in processing it — related
• Task-essentialness and productive use of new words in goal-directed activity may positively affect vocabulary learning and retention — a narrower finding that bears on this claim
• Answering history explanation questions often requires causal inferences because causal relationships are frequently left implicit in textbooks — related
• Learner-constructed graphic organizers are not shown to outperform provided ones: the one direct test, with college readers, favoured provided organizers on transfer — related
• Knowledge gained by self-analysis is more likely to produce constructive change in teaching than insights given by an observer — related
• Different evidence types differ in how strongly they can support claims about effectiveness — related
• AI-supported examination and construction practices (explanation, creation) qualify only as expanded or transformed epistemic practices because they can be performed without AI — related
• Faded steps were hypothesized to trigger specific self-explanation activities, tested via think-aloud protocols in Experiment 2 — related
• The meta-analysis coded and analyzed 20 moderator variables of the self-explanation effect — related
• Self-explanation prompting is potentially effective across a range of instructional conditions — possibly the same claim (merge candidate)
• Students who self-explain worked examples learn more effectively than those who do not, and the quality of self-explanations relates to explaining deep structure — a narrower finding that bears on this claim
• Self-explanation substantially increases time on task, raising questions about its efficiency — related
• Analogical reasoning improves transfer. — analogies are a core elaborative strategy
• Advance organizers improve learning. — organizers supply the prior-knowledge structure that elaboration connects to
• Revised post-teaching prompts (RPTAA) advanced candidates' capacity for more sophisticated analysis compared with the original prompts (PTAA) — a narrower finding that bears on this claim
• Language becomes an operational medium in AI-supported design, with risks of narrowing exploration — related
• Immediate Elaborated Feedback in a formative assessment game significantly improved students' programming knowledge, interest, and problem-solving versus systems without immediate feedback — related
• Learning through the experience of solving problems can build both content knowledge and thinking strategies — related
• Explanation prompts direct attention to targeted information and can reduce attention to other important information — related
• Self-explanation beats unguided studying for equal time, but instructional explanations, problem solving and retrieval practice can be equally or more effective — a narrower finding that bears on this claim
• Self-explanation promotes transfer of general patterns but can reduce recall of details and cause overgeneralization — a narrower finding that bears on this claim
• Self-explaining a worked example of a learning journal entry enhances strategy use and transfer of journal-writing benefits to a new topic — a narrower finding that bears on this claim
• Explicit training in constraint-seeking questions showed only moderate, non-generalizing benefits, whereas prompting explanation supports better question asking — a narrower finding that bears on this claim
• Self-explanation prompts in an instructional video do not improve retention or comprehension when learning time is controlled — reports the opposite
• Written explanation skills learned in the Unit 1 intervention transferred to a different mathematical concept (functions) — related
• Learning journal writing improves learning outcomes, but learners often do not apply sufficient strategies without support; prompts, worked examples and fading are the supports studied — related
• Non-interactive teaching is more than a special case of retrieval practice, with generative processing and social presence linked to its effects — a narrower finding that bears on this claim
• AI shifts the learner's role from producer to evaluator of mathematical meaning across multiple generative strategies — related
• Awareness prompts stimulated discourse about more concepts but did not improve shared drawing quality in elementary dyads — related
• Struggling to encode, retrieve, and reorganize information produces deeper comprehension and retention, and AI-supplied ready-made solutions short-circuit these processes — related
• Drawing improves STEM learning by a small-to-moderate amount over not drawing and over other active study strategies, in one meta-analysis — a narrower finding that bears on this claim
• Cognitive elaboration strategy significantly affects overall posttest performance, with self-generated elaboration highest — a narrower finding that bears on this claim
• Learning is an initiative construction of meanings completed by the interaction of learners' old and new knowledge — a broader claim this one bears on
• Adding a self-explanation feature to a base game improved transfer performance — a narrower finding that bears on this claim
• The effectiveness of each generative learning strategy is subject to potential boundary conditions discussed in the review — related
• Generative learning involves mentally reorganizing and integrating to-be-learned information with prior knowledge, enabling application to new situations — a broader claim this one bears on
• Prompting self-explanation in a physics electricity unit led to accelerated future learning in a subsequent magnetism unit — a narrower finding that bears on this claim
• Building extensive, well-organized domain knowledge may enhance functional use of intelligence in novel problems — related
• No evidence for a mapping between representational format and enhancement of a specific kind of understanding — related
• Anticipatory self-explanation prompts benefited higher-prior-knowledge apprentices on diagnostic strategy description scores, whereas retrospective prompts benefited lower-prior-knowledge apprentices — a narrower finding that bears on this claim
• Generative learning strategy prompts with metacognitive feedback yield higher self-regulation, strategy-use quality, recall and comprehension than strategy tools alone, controlling for prior knowledge — related
• Enforcing an explicit multi-step reasoning structure yields no significant improvement for the two studied models, which already self-organize an effective reasoning structure — related
• Receiving elaborate explanations benefits learning only when the receiver can apply the explanation in a related task — related
• Learners given appropriate materials and prompts can successfully develop, articulate, and elaborate rich epistemic ideals for good models and good sources — related
• Active-processing focus-support techniques such as testing, self-explaining, concept mapping, and drawing show positive effects on learning — a broader claim this one bears on
• Writing teaching scripts or textbook explanations improves university students' create-level research question generation more than writing study notes for their own learning — a narrower finding that bears on this claim
• Self-explanation benefits learning across ages, but evidence is mixed for elementary-school children and effects shrink when time on task is equated — a narrower finding that bears on this claim
• Drawing to learn shows positive effects on learning performance compared with non-generative and text-focused learning strategies — related
• Drawing can underperform the generative strategy of explaining, particularly on verbally aligned tests — a narrower finding that bears on this claim
• Provided visuals can act as a crutch: explaining with provided visuals yielded higher-quality explanations but no transfer advantage over explaining without them — related
• Visualizing support—provided diagrams or learner-generated drawings—facilitates higher-quality explanations and better comprehension than explaining text alone — a narrower finding that bears on this claim
• Test tasks requiring analysis, selection and organization of elements give better predictive results per unit cost than simpler tasks — related