Research

Teaching Geometry Online — what the research says, and how we built for it.

Geometry is the one school subject where a static diagram does the most damage: it arrives finished, hides the order it was built in, and tempts students to mistake a convincing picture for a proof. This is a plain-language synthesis of what the literature says works instead — and the case for Play With Geometry, 362 interactive ICSE geometry lessons for Classes 6–10, built in fifteen days around it.

Last updated June 2026 · Audience: teachers, parents, instructional designers.

The short version

Three findings dominate the geometry-education literature. Students learn a theorem by dragging a figure and watching what refuses to change; they climb the levels of geometric thinking by constructing figures with their own hands, not just viewing them; and they finally need proof only when it is sold as explanation rather than verification. Worked examples, productive failure, spacing, interleaving and step-level feedback all stack on top. Play With Geometry's five-stage loop is a direct read-out of that stack.

What we built — one lesson, five stages
See

The AI draws the syllabus figure live, line by line.

Break

Drag the handles — watch the theorem hold as everything else changes.

Build

Reconstruct the figure yourself with real tools, one validated step at a time.

Prove

Put the proof back together by ordering its steps.

Solve

Finish on a real ICSE board-pattern question.

The two ideas that shape Play With Geometry
Finding 1

Dragging a figure is how a student discovers what the theorem actually claims

Marton's variation theory holds that a critical feature of a concept becomes discernible precisely when it is the one thing held invariant while everything else varies. A dynamic figure makes this physical: grab a vertex, move it, and everything that can change does — but the property the theorem asserts stays fixed. The angle sum that refuses to leave 180° as you flatten the triangle is variation theory you can feel.

Research on dragging in dynamic geometry — Arzarello and colleagues' catalogue of dragging modalities, Baccaglini-Frank & Mariotti's "maintaining dragging" model, and Leung's work on the discernment of invariants in Educational Studies in Mathematics — finds that students who drag don't merely confirm a stated fact. They generate the conjecture themselves, arriving at a statement with a premise, a conclusion, and the conditional link between them.

The measured payoff is large. Chan & Leung's systematic review and meta-analysis (Journal of Educational Computing Research) reports a standardized mean difference around 1.0 for dynamic geometry over static instruction; a 2024 three-level meta-analysis (Ji, Guo & Song — 107 studies, 10,507 students) reports a moderate-to-strong effect of d ≈ 0.63.

In Play With Geometry
  • See draws the figure; Break hands over the drag handles. Dragging A, B, C while "∠A + ∠B + ∠C = 180°" stays put is the variation-theory contrast made literal — the invariant is the theorem.
  • Put a prediction prompt before the drag ("what will the angle sum do as the triangle flattens?"). Prediction sharpens attention to the invariant that follows.
  • Vary one thing at a time so the critical feature is separable from the noise — drag a single vertex, freeze the rest.
Finding 2

Constructing the figure — not just viewing it — moves students up the van Hiele levels

The van Hiele model (visualization → analysis → abstraction → formal deduction → rigor) is the most durable framework in geometry education, and its central claim is uncomfortable: the levels are driven by instruction, not age. A student stuck at "it looks like a square" hasn't matured too slowly — they have never been asked to operate on the properties.

Embodied-cognition research and a long tradition of ruler-and-compass work argue that constructing a figure — choosing the tool, placing each point, watching the constraint hold — is what turns a picture into a set of properties. As one review puts it, "by constructing models of geometric shapes and physically transforming them, children examine the changes that occur." Studies that pair construction with van Hiele-designed tasks report measurable level gains.

This is the Concrete → Representational → Abstract sequence in geometric form, an independently evidence-based progression: handle the figure, then schematise it, then reason about it in symbols.

In Play With Geometry
  • Build asks the student to reconstruct the figure with real tools — point, segment, compass, bisector, perpendicular, midpoint, angle. The midpoint tool that drops the exact middle of a chord is a property made operable instead of eyeballed.
  • Sequence the tools so each click encodes exactly one geometric move, and validate the step — not just the final picture.
  • Run a single lesson from the visualization level (See) up to deduction (Prove) so the level climb happens in one sitting and is visible to the learner.
Supporting research

Six further findings the literature converges on, each tied to a stage of the lesson loop.

Finding 3

Proof should explain, not merely verify — and dynamic geometry forces the issue

Dragging is so convincing that it creates a famous problem: once a student has watched the angle sum stay at 180° through a hundred triangles, they see no need to prove it. De Villiers' classic analysis of the roles of proof (verification, explanation, systematization, discovery, communication) is the way out. Stop selling proof as verification — the drag already supplied conviction — and sell it as explanation: not whether the angle sum is always 180°, but why it must be.

De Villiers argues specifically that, in a dynamic-geometry setting, the natural and meaningful entry point to proof is its explanatory function. The empirical phase and the deductive phase are not rivals; the first earns the question that the second answers.

In Play With Geometry
  • Order matters: See and Break should convince first; Prove then answers the "why" that conviction has made the student ready to ask.
  • Frame the proof prompt as an explanation ("why can the angle sum never be anything else?"), not a verification ("prove that the angle sum is 180°").
Finding 4

Proof comprehension is about grasping the global structure

Studies of secondary proof comprehension repeatedly find that students can recite individual steps yet miss how the steps hang together — the reasoning present in their oral explanations drops out of their written proofs, and the global structure (what researchers call encapsulation) is exactly what they fail to grasp. Level-spanning, structure-first proof strategies improve understanding more than line-by-line drilling.

Reconstructing a scrambled proof by putting its steps back in order targets that global structure directly. It is the geometric analogue of a Parsons problem: the student isn't asked to author every line from scratch, but to see why this step must precede that one.

In Play With Geometry
  • Prove presents the proof as cards to order — Given, the equal parts, the congruence/similarity rule, then the conclusion — so the learner reasons about structure, not handwriting.
  • Write each card as one move with its justification attached ("OA = OC — radii of the same circle"), so the logical spine is visible.
Finding 5

An animated construction beats a static diagram for novices

The worked-example effect (Sweller and successors) is one of the most replicated results in instructional research: novices learn more from studying a clear worked example than from solving a problem cold, because problem-solving from scratch spends working memory on search instead of schema-building. In geometry the textbook diagram is the worst case — it arrives finished, with every construction line already drawn and no trace of the order it was built in.

Drawing the figure live, element by element, turns the diagram back into a worked example: the learner sees which line came first and why, instead of decoding a static result.

In Play With Geometry
  • See animates the construction in textbook order rather than dumping the finished figure, so the diagram teaches its own derivation.
  • Keep narration to the objective; let the drawing carry the worked-example load.
Finding 6

Let them break it first — productive failure and boundary cases

Productive failure (Kapur) shows that, for conceptual understanding and transfer, a short bout of exploration before instruction primes learners to attend to the canonical method when it arrives. A meta-analysis of 166 comparisons reports a positive effect on conceptual understanding without sacrificing procedural knowledge — provided the exploration is always followed by a clear resolution.

Dynamic geometry gives this a sharp geometric form: drag the figure to its boundary. Push a triangle until the altitude leaves it, or drag a vertex until the circumcentre flies outside — the degenerate and obtuse cases are where the conditions of a theorem reveal themselves.

In Play With Geometry
  • Break is the exploration phase: hunt for the case that almost breaks the rule, then let the lesson close the loop with the deductive reason.
  • Constrain only what must be constrained — keep a triangle acute when the lesson needs its centres inside, but otherwise let the learner reach the edge cases on purpose.
Finding 7

Finish on a real question, and make practice spaced and interleaved

A lesson that ends at "now you understand" leaves the transfer untested. Retrieval and spacing research (a 2025 meta-analytic review in Educational Psychology Review) and Rohrer's randomised trials of interleaved versus blocked mathematics practice both find the same thing: mixing problem types and spreading practice over time lowers performance during practice but substantially raises delayed-test scores, because the learner has to choose the right tool, not just apply the obvious one.

Geometry rewards this especially — the hard part of a board problem is usually recognising which theorem applies, a discrimination that only interleaving trains.

In Play With Geometry
  • Solve closes each lesson with a genuine ICSE board-pattern question on the same theorem, so understanding is cashed out as a mark-earning answer.
  • Across lessons, resurface a theorem days later and interleave it with neighbours, so students practise choosing the theorem, not just executing it.
Finding 8

Online, keep it low-stakes with immediate, specific feedback

The 2025 npj Science of Learning systematic review of AI-driven tutoring in K–12 reports consistently strong effects, largest when feedback is step-level and specific rather than a final-answer verdict. Separately, studies of online mathematics learning find motivation and self-efficacy decline — and anxiety stays elevated — over weeks unless the design absorbs the risk with autonomy, low stakes, and a visible sense of progress.

A construction tool that validates each step ("this point isn't on the circle yet") and lets the learner undo freely is exactly the low-stakes, step-level feedback the evidence asks for.

In Play With Geometry
  • Validate the construction step, not just the finished figure, and name what's off rather than flashing "wrong".
  • Make every attempt cheap — unlimited undo, no streak-shaming, progress measured against the specific theorem mastered.

Geometry misconceptions worth targeting directly

Decades of research on student geometry errors surface the same small set of misconceptions. Each has a specific cure in the dynamic-geometry literature — and a lesson built around it.

MisconceptionWhat it looks likeResearch-backed remedy
The diagram is the proofTreats a single accurate figure as sufficient evidence; sees no need to prove what "obviously" looks true.Drag to a deceptive-looking case, then reframe proof as explanation (de Villiers).
Looks equal ⇒ is equalJudges lengths and angles by eye; assumes near-equal marks are exactly equal.Live measurements that update on every drag, so the eye is checked against the number.
A square is not a rectangleRefuses class inclusion — won't accept that a square has every property of a rectangle.Drag a rectangle until it becomes a square; its rectangle-properties never stop holding.
Angle sum depends on the triangleBelieves a bigger or more stretched triangle has a different angle sum.Drag A, B, C freely while the ∠A+∠B+∠C = 180° readout stays put.
An altitude is always insideAssumes every altitude (and the orthocentre) sits inside the triangle.Drag toward an obtuse triangle and watch the altitude and orthocentre leave it.
Perpendicular bisector = angle bisectorConfuses the two loci; expects them to coincide.Construct each one; show one is equidistant from two points, the other from two lines.
Scaling changes the anglesThinks enlarging a figure changes its angles, or that similar means "roughly alike".Similarity lessons: drag the scale factor k — sides scale, angles stay fixed, ratios hold.
Congruent and similar are the sameUses the two interchangeably; ignores the side-ratio condition.Side-ratio read-outs that show k = 1 (congruent) versus k ≠ 1 (similar).
A tangent can meet the radius at any angleDoesn't connect the point of contact to a right angle with the radius.Tangent–radius right-angle mark shown live as the tangent point is dragged.

Sources

Peer-reviewed papers and meta-analyses behind the claims above. Every link goes to the publisher or an open-access copy.

  1. Marton, F. (2017). What is made possible to learn when using the variation theory of learning in teaching mathematics? · ZDM Mathematics Education
  2. Leung, A. (2017). Variation and Mathematics Pedagogy · ERIC / MERGA
  3. Baccaglini-Frank, A. & Mariotti, M. A. (2010). Generating Conjectures in Dynamic Geometry: The Maintaining Dragging Model · Technology, Knowledge and Learning
  4. Leung, A. et al. (2013). Discernment of invariants in dynamic geometry environments · Educational Studies in Mathematics
  5. Chan, K. K. & Leung, S. W. (2014). Dynamic Geometry Software Improves Mathematical Achievement: Systematic Review and Meta-Analysis · Journal of Educational Computing Research
  6. Ji, Z., Guo, K. & Song, S. (2024). Effects of Dynamic Mathematical Software on Students' Performance: A Three-Level Meta-Analysis · Journal of Educational Computing Research
  7. van Hiele, P. M. (1957/1986). The van Hiele Model of Geometric Thinking (overview) · Charles University WDS Proceedings
  8. (2024). Enhancing Students' van Hiele Geometric Thinking Levels Through Geometer's Sketchpad · Qubahan Academic Journal
  9. de Villiers, M. (1990). The Role and Function of Proof in Mathematics · Pythagoras
  10. (2017). Engaging students in roles of proof · Journal of Mathematical Behavior
  11. (2022). Level-spanning proof-production strategies to enhance students' understanding of proof structure · International Journal of Mathematical Education in Science and Technology
  12. Sweller, J. (1988). Cognitive Load During Problem Solving: Effects on Learning · Cognitive Science
  13. Kapur, M. (2014). Productive Failure in Learning Math · Cognitive Science
  14. (2025). A Meta-analytic Review of the Effectiveness of Spacing and Retrieval Practice for Mathematics Learning · Educational Psychology Review
  15. Rohrer, D., Dedrick, R. F. & Stershic, S. (2015). Interleaved Practice Improves Mathematics Learning · Journal of Educational Psychology
  16. (2025). A systematic review of AI-driven intelligent tutoring systems (ITS) in K-12 education · npj Science of Learning

See it in action

Five lessons are free to play — one per class, 6 through 10. The rest of the 362-lesson atlas unlocks with a subscription.

Open Play With Geometry

Citation. EasyICSE.com (2026). Teaching Geometry Online — what the research says. Available at https://easyicse.com/research/geometry

Hold to talk

Subscription Status