Skip to content

Digital Learning

Principle · All principles
Evidence · 72 claims (35 for, 34 mixed, 3 against) · 81 studies (20 causal, 20 review, 12 quant-synthesis, 12 theoretical, 6 associational, 6 design, 5 qualitative), q1–q4 · 13 of 81 report an effect size · 61 claims rest on one study

Conditional relationship

Digital learning is a medium: devices, platforms and networks. It is not a method. This page covers when to put part of a course onto a digital tool, and how to credit the result. The relationship it proposes is conditional. Suppose a course needs a function that cannot be supplied well enough face to face: more practice items with feedback on each one, adaptive selection of the next task, replay of an explanation, access across time and place, or a record of every attempt. Moving that function onto a digital tool that carries a named method (practice with feedback, contingent hints, prompts to explain, visible progress against a criterion) is then expected to raise outcomes on an aligned measure (general-achievement, immediate-performance). The comparison is the same course without that function. Moving the same method from paper or a classroom to a screen is not expected to change learning by itself. Any gain should be credited to the method and to the extra practice, time or feedback the medium made affordable. The medium gets credit for making those affordable, not for the learning.

One claim in the wiki tests a digital medium against a non-digital one with the method held equal, and it found no difference; none shows equivalence. Triona and Klahr (2003) taught 92 fourth- and fifth-graders to design unconfounded experiments with the same instruction on physical ramps or on a computer simulation. Both groups gained by the same amount (no effect of condition, F = .01), and a week later the children trained on the computer designed as many good experiments with the physical ramps as those trained on them. That is one study of one science skill, and a nonsignificant difference is not a test of equivalence. The rest of the evidence points at confounds. In a meta-analysis reported second-hand, computer-based instruction's advantage over conventional teaching fell from .51 SD to .13 when the same teacher taught both versions. In the 2009 US Department of Education meta-analysis, online-versus-face-to-face effects were +0.20 where curriculum and instruction were judged near-identical and +0.42 where they differed. A systematic review of immersive virtual reality comparisons found 40% confounded by instructional method or content, and where learning activities were not controlled the confound favoured the VR arm in 14 of 17 comparisons. Intelligent tutoring systems beat other computer-based instruction (g = 0.57): same medium, different method. Blended courses beat face-to-face ones (g+ = 0.35), but with extra time and resources in the blended arms. A methodological synthesis argues that whole-medium comparisons confound medium with method, novelty and population, and that their usual "no significant difference" result is uninterpretable, not proof of equivalence. The steps below are therefore a default design. They follow the method claims, not the medium.

Several converted pages own narrower relationships that sit inside this one, and this page does not repeat them:

The Blended Learning pattern holds how to divide a course between online and in-person work, and flipped learning is one configuration inside it. This page decides whether a digital tool earns its place and what it is for. Those pages decide how the method runs.

Default design, while the relationship is untested

The earlier page's guidance is kept here as a concrete default, and each step states its evidence status. Durations and thresholds are untested proposals to set locally and record.

  1. Name the method before the tool. Write one line for each digital component: "this tool lets learners do X (method) n times a week, which we cannot supply in class because Y". Drop any component for which no X or Y can be written, for example a lecture posted as a video with nothing asked of the learner, or a worksheet turned into a PDF. Basis: Media comparison studies produce uninterpretable "no significant difference" findings. [+M] and The measured advantage of computer-based instruction over conventional teaching shrinks when the same teacher teaches both versions [+M], which argue that the medium alone does little. A randomised evaluation of One Laptop per Child in rural Peru, reported second-hand, raised computers per student from 0.12 to 1.18 and found no effect on mathematics or language scores after fifteen months (Distributing hardware without capable software produced no measurable effect on mathematics or language scores in a large randomized evaluation, as the article reports citing Cristia et al. [+W]): devices with no named method attached. The one-line test itself is an untested proposal.
  2. Use the medium for practice with feedback on each item. Short sets of 8–15 items, about 10–20 minutes, 2–4 times a week. Show the correct answer and a reason after each item, or after the set where Immediate Feedback argues for a delay. Do not cut practice out of an online lesson to save time. Basis: In one experiment with 256 undergraduates, computer-based lesson versions that included practice produced significantly higher posttest scores than versions without practice [+M] (one computer-based lesson; horizon not reported). The set sizes and frequency are untested proposals.
  3. Ask for explanations, not only answers. Add one "why does this step work?" or "what principle applies here?" prompt for every three to five items. Allow a short typed answer or a choice among explanations, and fade the prompts once a learner's explanations are accurate. Basis: Prompting learners to self-explain improves understanding and problem solving on immediate tests by a moderate average amount, with little evidence yet on delayed or classroom gains. [+S] (a meta-analysis across task types and levels, g = .55). It is not specific to digital delivery: the prompt does the work, whatever medium carries it. Inside one tutoring system, a classroom experiment found that students asked to explain their steps in the Geometry Cognitive Tutor gained more on reason-giving and "not enough information" items than students who only solved problems (Static self-explanation prompts in the Geometry Cognitive Tutor produced greater gains on Reason and Not Enough Info test items than problem-solving practice alone [+M]; no effect size printed).
  4. Give hints that respond to the attempt. After an error, give a cue first, then a more specific hint, then the worked step. Give less help after a success. Where a tutoring system is available for the domain, it bundles this with practice and a model of the task. Basis: Contingent scaffolding improves learning more than fixed or absent support. [~M] (mostly one-to-one human tutoring; carried here to software) and Intelligent tutoring systems improve achievement over non-individualised instruction [+S] (intelligent tutoring beat teacher-led large-group instruction, other computer-based instruction and textbooks, but not human tutoring or small groups). A meta-analysis of 144 experiments found students given computer-based scaffolds in STEM did better on cognitive tests than those without them, with wide variation between studies (Computer-based scaffolding shows a consistently positive effect on cognitive outcomes in STEM education [+M]). Within one tutoring system, adding real-time feedback to the same adaptive diagnostic tutor raised progression on a SOLO-scored mathematics assessment (d = 0.73, reported in a review; Automated real-time feedback in an adaptive diagnostic ITS produced greater progression on a SOLO-based mathematics assessment than the same system without real-time feedback [+W]). Which part of the bundle does the work is otherwise not shown. If a generative-AI tutor carries the hints, set it to withhold final answers and give hints instead. In a trial reported second-hand by several papers, unrestricted access to a general chatbot raised practice scores by 48% and left students 17% worse on the unaided exam once it was removed, while a tutor rebuilt to withhold answers removed most of that loss (Review reports a field experiment in which GPT access improved supported practice but was followed by poorer unaided test performance, mitigated by a guarded tutor [~M]). The model was the same in both arms; what learners were asked to do with it changed the result.
  5. Show progress against a stated criterion, and ask the learner to act on it. A dashboard of completed items is not enough. Show accuracy on each skill against a target, and once a week ask "what will you practise next, and why?". Basis: Self-regulated learning models treat self-monitoring as a core process through which learners notice gaps and adjust strategy [~M] (a review and a theoretical synthesis; no test of dashboards).
  6. Keep the interface and the materials lean. Use one place to find the week's work, the same layout each week, and no decorative media. Segment long videos into parts of a few minutes, each with a question after it. Basis: Cognitive Overload Degrades Learning [~M] (reviews of multimedia design: cutting extraneous load helped most on complex material) and Segmenting Improves Multimedia Learning [+M] (the effect held for system-paced segments). A review of learner-controlled hypermedia names disorientation, distraction and cognitive overload as the usability problems that make its benefits hard to show (Usability problems — disorientation, distraction, and cognitive overload — hinder demonstrating effectiveness of self-controlled hypermedia learning [~W]; an argument, no effect sizes). On video, a review reports that Guo, Kim and Rubin found short videos more engaging and classroom recordings weak online (Guo et al. found engagement depends on video production style: short videos, talking heads on slides, Khan-style drawing more engaging than slides or coding, classroom recording weak online [+W]; engagement, not learning, and edX science and engineering courses only), and that including the instructor's face showed no significant learning-performance difference, though some learners prefer it (Human embodiment in video: perceived social presence benefits learning ratings, but instructor-face inclusion shows no significant learning-performance difference, and learners prefer human over robot presenters with mixed recall [~W]; second-hand, equivalence not tested). So a talking head is a preference to offer, not a learning requirement. Interface load as such is untested here.
  7. Make the first week a contact week. Most of the extra attrition online in the one study recorded here came before instruction began. In week 1, give a short task that needs a reply from a person (a post the instructor answers within 24–48 hours, or a 15-minute live call), and check that every learner has logged in and completed one item by day 3. Basis: Online attrition during Orientation Week is twice that of onground classes' first week [~M] (university continuing education, observational). The contact task and its timings are untested proposals.
  8. Check access before launch. Find out who has which device, what bandwidth, and what accessibility needs. Provide an offline or printable route for each required activity, and do not grade on anything that some learners cannot reach. This step is an untested proposal, kept from the earlier page. The condition it answers is documented: national survey figures put individual internet use in rural Bangladesh at about half the urban rate (Individual internet penetration in rural Bangladesh (36.5%) is about half the urban rate (71.4%), which one case-study preprint argues puts cloud-dependent AI tools out of reach for rural learners [~W]; one country, and an offline route was not tested).
  9. Plan the arc across weeks.
  10. Weeks 1–2: the contact task, items close to a worked example, the full hint ladder, and feedback after every item.
  11. Weeks 3–6: hints on request only, a self-check before submitting (compare with a worked answer, or check an inverse), and explanation prompts reduced to one per set.
  12. Final weeks: mixed items in the form of the target assessment, no hints, feedback after the set.
  13. Open each week with three or four unaided items on the previous week's material. Move a learner to the next unit, or to extension work, once they reach about 80% on two consecutive unaided sets, so that learners who are already accurate are not held back (Instructional guidance that helps novices can become redundant or counterproductive as expertise grows. [~M]).
  14. The arc, the 80% figure and the two-set rule are untested proposals.
  15. Evaluate the method, not the medium. Do not report "digital versus traditional". Compare two versions of the same digital component that differ in one thing (with or without explanation prompts, immediate or end-of-set feedback). Or compare the component with the same method on paper at the same time on task. Measure unaided performance at least once after a delay. If a tool is compared with a conventional version, write down the activities each arm does before the results come in, because unequal activities inflate the newer medium's apparent advantage. Basis: Media comparison studies produce uninterpretable "no significant difference" findings. [+M]; Effect sizes were larger in studies where online and face-to-face conditions differed in curriculum materials and instructional approach (+0.42) than where curriculum and instruction were nearly identical (+0.20; Q = 5.40, p < .05). [+M]; 40% of IVR-vs-conventional comparisons had at least one confound related to instructional method and content [+M] and When activities were not controlled, the confound favored the IVR condition in 14 of 17 comparisons [+M] (a systematic review of 50 STEM comparisons). The specific evaluation plan is a proposal.

When a condition is not met: if no function can be named that the classroom cannot supply, keep the activity off-screen. If access is uneven and cannot be fixed, make the digital route optional and equivalent, not required.

Fitting the design to a situation

The three facts that most change the decision:

  • What the digital tool will do that cannot be done otherwise. More practice with feedback, adaptive selection, replay, or reach to learners who cannot attend all justify it. "Engagement" or "modernising" do not. If the brief omits it, ask: which activity gets more attempts, more feedback or more access because it is on a screen, and how many more?
  • Whether a person is in the loop, and how quickly. Fully self-paced work with no human contact loses learners early and puts the whole burden of pacing on them. If the brief omits it, ask: who will see a learner's work or absence in the first week, and how soon will they respond?
  • The learners' access and their starting response on the task. Devices, bandwidth and accessibility needs decide what can be required. Prior knowledge decides how much guidance the software should give. If the brief omits it, ask: can every learner reach the platform on the device they have, and can they do one item of the target kind unaided now?
If the brief says… Then change… Basis
Learners are novice on the task Use items close to worked examples, the full cue–hint–worked-step ladder, an explanation prompt every three to five items, and feedback on every item Contingent scaffolding improves learning more than fixed or absent support. [~M] (one-to-one tutoring, carried to software)
Learners are advanced on the task, or succeed unaided on the first set Remove most hints and prompts; give mixed, whole-task items and feedback only on request Instructional guidance that helps novices can become redundant or counterproductive as expertise grows. [~M] (a cognitive-load review; gives no crossover point)
Learners are child (primary age) Use short sessions (10 minutes or less), supervised by an adult in the room; give feedback that needs little reading; avoid open typed answers. A simulation can stand in for physical materials if the same instruction goes with it session length and supervision are untested proposals; the substitution rests on Physical and virtual instructional materials are equally effective for teaching fourth- and fifth-graders to design unconfounded experiments when instructional method is controlled [+M] (one science skill; no difference found, equivalence not tested). Online-versus-classroom comparisons for school-age learners are almost absent: A systematic search found almost no rigorous studies contrasting online and face-to-face instruction for K–12 students: none with computable effect sizes through 2006, and only five meeting meta-analysis criteria through July 2008. [~M]
Learners are adult or professionals studying around work Use short sets on a phone that can be done in one sitting, and tasks taken from their own work; keep the first-week contact task untested proposal; contact task from Online attrition during Orientation Week is twice that of onground classes' first week [~M] (university continuing education)
Goal is a procedure or verbal-association with checkable answers Put the bulk of practice online with automatic checking; use an intelligent tutor if one exists for the domain Intelligent tutoring systems improve achievement over non-individualised instruction [+S] (tutoring-system meta-analyses; gains smaller on standardised than on locally developed tests)
Goal is a concept, a principle, or transfer Keep answer-checking, but add explanation prompts and changed items; hold discussion of reasons in a live or threaded session Prompting learners to self-explain improves understanding and problem solving on immediate tests by a moderate average amount, with little evidence yet on delayed or classroom gains. [+S] (not specific to digital delivery)
Setting is online-self-paced, no teacher present Add the contact task in week 1, a fixed weekly rhythm with a deadline, visible progress against a criterion, and a weekly plan prompt; check logins by day 3 Self-regulated learning interventions have a moderate effect (0.65) on learning outcomes in online and blended environments [+M] (15 studies); a diary alone did not help in one course (A daily learning diary alone (Group D) did not produce statistically significant pre-post gains on any measured outcome in an online mathematics preparation course [~M])
classroom, one teacher and a large group, devices available Use the software for individual practice and hints while the teacher watches the class's results live and goes to those who are stuck; use class time for what the software cannot do (discussion of reasons, whole tasks) untested proposal; contingent-support evidence is one-to-one (Contingent scaffolding improves learning more than fixed or absent support. [~M])
Time is short (one or two sessions) Do not introduce a new platform; use the classroom, or one familiar tool, for practice with feedback untested proposal; learning a new interface costs time the method needs
Note-taking during lectures on laptops or tablets is planned Ask learners to summarise in their own words, not transcribe; do not ban devices on the strength of the longhand evidence Laptop note-takers transcribed more verbatim and did worse on conceptual questions than longhand note-takers in one set of experiments, but a direct replication found no consistent difference in test performance [~M] (college students; a replication found no consistent difference)
Stakes are high, or the final assessment is unaided and supervised Judge readiness only on unaided items done without hints; keep a log of hints used; give at least one supervised check in the final assessment's format untested proposal; its reasoning is Salomon's distinction between what a learner does with the tool and what remains when working alone (Salomon distinguishes effects with the computer (system performance) from effects of the computer (cognitive residue on the solo performer) [+W], a conceptual argument reported second-hand)
Access is uneven (shared devices, poor bandwidth, accessibility needs) Make every required activity available offline or on paper, use low-bandwidth formats (text, short audio, captioned video), and grade nothing that some learners cannot reach. Check that the software runs on the phones learners own, since out-of-class work may otherwise depend on finding a computer untested proposal, from the earlier page; the condition is documented in Individual internet penetration in rural Bangladesh (36.5%) is about half the urban rate (71.4%), which one case-study preprint argues puts cloud-dependent AI tools out of reach for rural learners [~W] (survey figures for one country; no offline design tested) and, for adult education, Learners used products outside class despite infrastructure limits: 65% of surveyed students reported doing software work outside class while mobile access was largely unavailable [~W] (a survey at fourteen sites; smartphone ownership figures are second-hand)
A generative-AI chatbot or tutor is part of the plan Use a tutor set to withhold final answers and give graded hints, not a general chatbot; judge readiness only on unaided items with the tool removed, both straight after and a week later; ask learners to check its output, and record errors it makes Review reports a field experiment in which GPT access improved supported practice but was followed by poorer unaided test performance, mitigated by a guarded tutor [~M] (one high-school mathematics trial, reported second-hand); a structured tutor raised post-test scores over a no-tutor arm in one small preregistered field experiment (Access to a structured conversational AI tutor increased post-test content scores over a no-AI holdout with ambient consumer AI available [+M]; 46 analysed). Unrestricted access in a 35-minute learning phase raised unaided scores straight after (0.28 SD) and by 0.21 SD a week later, not significant (AI access raises immediate unaided test scores by 7.2 percentage points (0.28 SD) after a timed learning phase [~M], One week later, the AI-access group's lead on unaided assessments shrinks to 3.9 pp (0.21 SD), short of statistical significance (p = 0.068) [~M]), so unrestricted access does not always harm; the configuration and the task decide. Errors and lack of real-time response were the weaknesses learners named in one music study (Students identified weaknesses in the AI feedback model: lack of real-time feedback, software/economic access barriers, and occasional incorrect or inconsistent outputs [~W])

Seven of the thirteen rows rest on a claim. Five more cite one only for part of the move (the adult row's contact task, the child row's substitution of a simulation for physical materials), for its limit (the classroom row), for its reasoning (the stakes row) or for the condition it answers (the access row). Three of the seven were tested in digital settings: self-regulation interventions online and blended, laptop note-taking, and generative-AI tutors, whose trials are few, short and mostly reported second-hand. The tutoring-system row was tested on software, but the software bundles several methods. The others carry a method claim from tutoring, cognitive-load research or self-explanation studies into a digital setting and say so. The short-time row is an untested proposal.

Observation, state and explanation

What can be observed in a digital setting: logins and their times, items attempted, answers, time per item, hints requested and at what level, which explanation prompts were answered and how, and drops. "Disengaged", "not self-regulating" and "the platform works" are inferred states. Logs record what the software saw. A long time on an item can mean thinking, a distraction or a closed tab.

Response Plausible explanations Observation that could change the interpretation
Learner never logs in after enrolling, or stops in week 1 Access problem (device, bandwidth, password); unclear what to do first; no human contact yet; competing commitments Contact the learner directly in week 1. Ask whether they could reach the platform and what they would need, and record the reason.
High scores on practice items, low on the unaided check Hints carried the answers; the items were too close to worked examples; answers were looked up or guessed; an AI assistant did the work (the pattern in the GPT-4 trial [~M]) Compare accuracy on items done with no hints against those with hints, and give a changed item with no hints or AI available. Ask the learner to explain one practice answer in their own words.
Many attempts in quick succession on the same item Guessing through the options; misunderstanding of the feedback; a habit carried over from games Delay feedback until an explanation has been typed, then check whether accuracy changes. Ask what the feedback meant.
Completes everything, but answers explanation prompts with one word or a copy of the question Prompt seen as busywork; does not know what a good explanation looks like; language or typing barrier Show one model explanation, offer a choice among explanations, or accept a voice note; compare the quality afterwards.
Outcomes better than last year's face-to-face cohort More practice or time on task; a different cohort; a different test; a novelty effect; a real method effect Compare time on task and the test used; run the same items on paper with a matched group; re-check after the novelty has worn off.

These rows are proposals and have not been tested with learners as diagnostics. A probe changes what it measures, so record each one, and keep the explanations that remain open rather than choosing the one whose name fits.

Evidence and qualifications

One claim compares a digital and a non-digital version of the same method and finds no difference; none shows equivalence. The claims below bear on the attribution rule, and on what methods do inside the medium.

The learners, outcomes and dates differ widely: college students in the 1970s, physics undergraduates, continuing-education enrolments, tutoring-system studies across levels. So do not rank these by effect size. None measures delayed-retention of a digital-versus-non-digital difference.

The evidence by question

Further evidence, not yet read against this model

Claims read against this model that bear on it only at the edges, or not at all. The other claims once listed here, and those the linker added on 2026-10-15 to 2026-10-19, are now cited in the steps, rows and evidence above.

Objective and learner-valued goal

The designer's objective is usually unaided performance on a representative task at a stated horizon. Ask separately what learners want from the digital part. Some want to fit study around work or care, some want to replay what they missed, some want to avoid being seen getting things wrong, and some want simply to finish the module. Agreement and divergence are both observations. A learner who values flexibility may welcome self-paced sets and drop out without a fixed weekly rhythm. One who values contact may experience a platform with no person in it as being left alone.

For example, a nurse in a continuing-education course may want something to check doses on during a shift, while the designer's objective is unaided calculation accuracy at the end of the module. The software practice serves the objective. A short reference card with worked examples, used and then removed for the unaided check, serves the learner's goal without letting assisted success stand in for the capability. Record both, so that completion rates are not read as the capability either party wanted.

What would revise this model?

The model should change in five cases:

  • If studies that hold method, time and materials equal and are powered to detect a modest difference find a consistent effect of the medium itself (in either direction), the medium should be credited, and "medium, not method" should become a stated exception. The one such study in the wiki (Triona and Klahr, physical against virtual materials) found none; an equivalence test with a stated margin would settle more than another nonsignificant difference.
  • If generative-AI tutors set to withhold answers fail to protect unaided performance across further trials, or unrestricted access consistently helps delayed unaided performance, the generative-AI row changes.
  • If the same digital method performs consistently worse than its paper or classroom version (as one physics comparison hints) for reasons the interface explains, interface load becomes part of the model, not a footnote.
  • If the first-week contact task does not reduce early attrition when tried, drop it and look for the cause in access or enrolment.
  • If tutoring systems' gains vanish on standardised and delayed measures, the case for software-delivered adaptive practice narrows to aligned, immediate outcomes.

Do not protect the model by calling every failure "poor implementation" after the fact. Record the configuration before the results.

Access, use, assisted success, unaided immediate performance, delayed retention and transfer are separate outcomes. The present evidence establishes no dose of digital practice, no threshold for moving on, and no effect of the medium itself.

Claims

  • Flipped Learning — digital delivery often makes pre-class access and replayable initial exposure possible
  • Learner Choice — platforms can support alternative pathways, modalities, and pacing
  • Multimedia Learning — digital environments frequently instantiate multimedia principles in practice
  • Adaptive Learning — owns choosing each learner's next task from their responses, the main reason to put practice on software
  • Immediate Feedback — owns when feedback should arrive; software makes it immediate by default, which is a choice to make, not a given
  • Self-Regulated Learning — owns the pacing and persistence that self-paced digital work demands
  • Cognitive-Load Management — interface and presentation load in digital materials

Examples

Illustrative

Flipped Learning — Learners access short digital explanations before class, then use live time for practice, discussion, and feedback. The gain comes from redesigning class time, not just posting videos.

ASSISTments — A digital math platform that combines practice, hints, and immediate feedback. It shows the value of digital learning when the environment supports reasoning and feedback rather than just answer submission.

Shared Documents — Learners jointly annotate, revise, and synthesize ideas in a shared digital workspace, turning the platform into a medium for collaboration rather than just content storage.

Blended Learning — the pattern for dividing a course between online and in-person work, so that each mode does what the other cannot.

Key Sources

  • Means, B., Toyama, Y., Murphy, R., Bakia, M., & Jones, K. (2010). Evaluation of evidence-based practices in online learning. U.S. Department of Education.
  • Hodges, C., Moore, S., Lockee, B., Trust, T., & Bond, A. (2020). The difference between emergency remote teaching and online learning. EDUCAUSE Review. https://er.educause.edu/articles/2020/3/the-difference-between-emergency-remote-teaching-and-online-learning
  • Qi, G. Y., & Wang, Y. (2022). Challenges and responses: A Complex Dynamic Systems approach to exploring language teacher agency in a blended classroom. The JALT CALL Journal, 18(1). https://doi.org/10.29140/jaltcall.v18n1.569
  • V.O.M. Lara, D. F. Amaral, D. Faria, and L. P. Vieira. (2014). Demonstrations of magnetic phenomena: Measuring the air permeablity using tablets. https://arxiv.org/abs/1405.6581
  • Triona, L. M., & Klahr, D. (2003). Point and Click or Grab and Heft: Comparing the Influence of Physical and Virtual Instructional Materials on Elementary School Students' Ability to Design Experiments. Cognition and Instruction, 21(2), 149–173. https://doi.org/10.1207/s1532690xci2102_02
  • Lawson, A. P., Martella, A. M., LaBonte, K., Delgado, C. Y., Zhao, F., Gluck, J. A., Munns, M. E., LeRoy, A. W., & Mayer, R. E. (2024). Confounded or Controlled? A Systematic Review of Media Comparison Studies Involving Immersive Virtual Reality for STEM Education. Educational Psychology Review. https://doi.org/10.1007/s10648-024-09908-8
  • 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
  • Means, B., Toyama, Y., Murphy, R., Bakia, M., & Jones, K. (2009). Evaluation of Evidence-Based Practices in Online Learning: A Meta-Analysis and Review of Online Learning Studies. U.S. Department of Education.
  • Wee Loo Kang. (2013). Open Source Physics. i in Practice 1(1), 58-63. http://ictconnection.opal.moe.edu.sg/cos/o.x?ptid=711&c=/ictconnection/ictlib&func=view&rid=82
Page metadata
Field Value
type principle
id digital-learning
aliases shared-whiteboard-dual-cohort-workspace, tablets-expand-science-education-approaches, open-licensing-teachers-as-designers
title Digital Learning
description Putting part of a course on a digital tool is expected to help only through the method it makes affordable (more practice with feedback, adaptive hints, prompts to explain, visible progress) and not through the medium itself; no claim here tests a digital against a non-digital version of the same method.
status review
canonical True
generated by claude/unspecified on 2026-10-19
trust unverified — no one has confirmed this page's content
sources 4 — hodges-2020, qi-2022, vom-lara-2014, wee-loo-kang-2013

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