Math manipulatives research: hands-on, with conditions.

Walk into any early-years classroom and count the objects that exist to teach number. Interlocking cubes. Bead strings. Base-ten blocks. Fraction circles. Plastic bears in three sizes. The global market for these materials is worth serious money, and their presence has become a proxy for good practice: an inspector who sees children moving blocks is inclined to see learning. The intuition behind the spend is old and respectable — Pestalozzi, Montessori and Piaget all argued, in different registers, that young children think through things before they think through symbols.

The intuition is not wrong. But for most of a century it went largely untested at the level that matters: do children taught with objects end up better at the mathematics — not the objects — than children taught the same content with symbols alone? When researchers finally forced that comparison at scale, the answer came back conditional. Sometimes yes, sometimes no, sometimes backwards. The interesting part of the literature is not the average. It is the conditions that flip the sign.

This article walks the evidence in three moves. First, the meta-analysis that put a number on the average and identified the moderators. Second, the cognitive account — objects as symbols — that explains why concrete help is never free. Third, the design rule that the best experiments now support: concreteness fading, the deliberate journey from thing to notation. The destination is a working answer to the question every teacher actually has: not “are manipulatives good?” but “when do I bring them out, and when do I take them away?”

The intuition and the industry

Start with why the belief runs so deep. Children plainly engage more with things than with worksheets. Objects give wrong answers a visible shape: a child who thinks 23 minus 8 is 25 can watch the blocks disagree. And the developmental story feels right — number names are abstractions, and a fistful of cubes seems like the natural bridge to them. By the 1990s, hands-on maths was official orthodoxy in many systems, written into standards documents and inspection rubrics alike.

Two things should have prompted more caution. The first was the state of the evidence: classroom studies existed, but they varied wildly in quality, duration and what they measured, and reviewers kept returning verdicts of “mixed”. The second was a quiet finding from developmental psychology: the ability to use an object as a symbol for something else is late-arriving, fragile machinery — not a freebie of childhood (DeLoache, 1987). The field bought the bridge before checking the load rating. The next two sections are the check.

What the meta-analysis found

The modern anchor is Carbonneau, Marley and Selig’s meta-analysis: 55 studies, several thousand students from preschool to college, all comparing maths instruction with concrete manipulatives against instruction relying on abstract symbols alone (Carbonneau, Marley & Selig, 2013).

The average result was positive but modest — on the order of a third of a standard deviation. That is a real instructional effect, worth having, and far short of the transformation the catalogues imply. More useful than the average was its spread. Effects were larger when researchers measured retention of what was taught, and small for the outcomes schools care most about long-term: transfer and problem solving (Carbonneau, Marley & Selig, 2013). Objects, used typically, help children hold the taught procedure. They do far less to help children carry the idea to new territory.

The moderators carried the practical news. Effects depended on the level of instructional guidance — manipulatives embedded in structured teaching outperformed manipulatives as free exploration. They varied with the perceptual richness of the objects themselves. And they varied with age, with younger children showing the pattern the symbol account predicts: the bridge is hardest to cross for exactly the children it is bought for (Carbonneau, Marley & Selig, 2013). A tool whose effect ranges from helpful to harmful depending on use is not a purchasing decision. It is a design problem.

Objects are symbols: the double demand

Why would a device this intuitive deliver results this uneven? The best answer starts in a study with no maths in it at all. Judy DeLoache showed young children a scale model of a room, hid a miniature toy in the model, and asked them to find the real toy in the real room. Three-year-olds walked straight to it. Children just six months younger failed — while remembering perfectly where the miniature was hidden (DeLoache, 1987). The younger children saw the model vividly as a thing. What they could not yet do was see it as a thing and a symbol for something else at the same time.

DeLoache called that ability dual representation, and Uttal, Scudder and DeLoache used it to reframe the entire manipulatives debate: a block standing for a hundred is exactly such a double object (Uttal, Scudder & DeLoache, 1997). A child using base-ten blocks is being asked to perceive the wood and, simultaneously, to look through the wood to a quantity relation. Nothing about touching the block guarantees the looking-through. A child can become genuinely expert at the material — stacking it, trading ten of these for one of those — while the intended mathematics never arrives. The manipulation is visible; the representation is not.

The account makes a sharp, testable prediction. The more interesting an object is as an object, the harder the double demand becomes — attention sticks on the surface. So the toy-like materials chosen to be engaging should, on this theory, be precisely the ones that teach the least. That prediction has now been tested many times, and it keeps winning (Uttal, Scudder & DeLoache, 1997).

When concrete hurts

Two experiments made the cost concrete, so to speak. McNeil and colleagues gave students money word problems accompanied by either perceptually rich play money — realistic bills and coins — or bland paper versions. The rich materials, the engaging ones, went with more errors on the target problems. The authors’ title carried the verdict: concrete objects both hurt and help, with the richness doing much of the hurting (McNeil, Uttal, Jarvin & Sternberg, 2009).

The second experiment chased the deeper outcome — transfer. Kaminski, Sloutsky and Heckler taught learners a simple mathematical structure through either concrete, story-rich instantiations (measuring cups of liquid, among others) or a single generic symbol system. Everyone learned their training version. Then came a novel task built on the same underlying structure. The generic-symbol learners applied the structure well — roughly three-quarters correct. The concrete learners performed near chance (Kaminski, Sloutsky & Heckler, 2008). The vivid version was easier to like and harder to leave: what was learned stayed welded to cups and liquid.

Neither study says concrete materials are bad. Grounding has real benefits — the money helped some students set problems up; stories give meaning that bare notation lacks. What the studies kill is the assumption that concrete is free. Every degree of vividness buys engagement and immediate sense-making, and charges for it in attention and portability (McNeil, Uttal, Jarvin & Sternberg, 2009). Once you see the trade, the design question stops being “objects or symbols?” and becomes “in what order, and who pays the exit cost?”

Concreteness fading: the design rule

If concrete grounds meaning and abstract travels, the obvious move is to refuse the either/or. Sequence them. That idea has a long pedigree — Jerome Bruner’s enactive–iconic–symbolic progression sketched it in the 1960s — and a modern name: concreteness fading (Fyfe, McNeil, Son & Goldstone, 2014).

The canonical version has three steps. Start concrete: physical or richly depicted material that lets the learner see what the mathematics is about. Then fade: a stripped, schematic form — pictures of the blocks, then marks standing for them — that keeps the structure and sheds the surface. End symbolic: bare notation, the format in which the knowledge must eventually live. Goldstone and Son supplied the first strong evidence in science simulations: learners who started with concrete elements that faded into idealised forms transferred principles better than learners held at either extreme (Goldstone & Son, 2005).

Fyfe, McNeil, Son and Goldstone’s systematic review then pulled the maths and science strands together. Across experimental comparisons, the faded sequence tended to beat concrete-only, abstract-only, and — importantly — the reverse order, with the benefit clearest for novices and for transfer measures (Fyfe, McNeil, Son & Goldstone, 2014). Direct classroom-style experiments back the specific claim. In studies of children learning maths equivalence, fading produced better transfer than the same materials arranged in any other order — including abstract-to-concrete, which shares every ingredient and differs only in direction (Fyfe, McNeil & Borjas, 2015).

That last comparison is the one worth memorising, because it isolates the principle: the power is not in the blocks, and not in the notation. It is in the path between them. Manipulatives, in this reading, are not a method. They are the first third of a method, and their value is realised only at the moment they are taken away (Fyfe, McNeil & Borjas, 2015).

What the evidence doesn’t show

This literature supports a design rule, not a slogan in either direction. Six limits keep the reading honest.

• The average is almost meaningless alone. The pooled ≈0.37 SD spans studies whose effects run from strongly positive to negative depending on guidance, object design, outcome and age — quoting the mean without the moderators misrepresents the meta-analysis (Carbonneau, Marley & Selig, 2013).
• The sharpest experiments are the least classroom-like. The famous transfer failures used adult learners, minutes-long training and novel structures; how far they stretch to seven-year-olds across a school year is genuinely contested (Kaminski, Sloutsky & Heckler, 2008).
• Fading’s evidence base is young and narrow. The strongest direct tests cover a handful of topics — maths equivalence, simple structures, science simulations — with short delays; long-term, curriculum-scale trials are still scarce (Fyfe, McNeil, Son & Goldstone, 2014).
• Stage lengths are unparameterised. No study yet tells a teacher how long to stay concrete or when precisely to fade; current practice guidance interpolates between conditions rather than reading off a dose-response curve (Fyfe, McNeil & Borjas, 2015).
• Physical versus virtual is unsettled. Object type moderates effects in the meta-analytic record, but clean comparisons of physical blocks against their on-screen equivalents remain too few and too mixed to call (Carbonneau, Marley & Selig, 2013).
• None of this bans objects. Concrete grounding measurably helps problem set-up and meaning-making — the same money study that found richness hurting also found concreteness helping — so the finding is about sequencing and exit, not prohibition (McNeil, Uttal, Jarvin & Sternberg, 2009).

Where the evidence stops

1. 1The average is almost meaningless alone
2. 2The sharpest experiments are the least classroom-like
3. 3Fading’s evidence base is young and narrow
4. 4Stage lengths are unparameterised
5. 5Physical versus virtual is unsettled
6. 6None of this bans objects

© 2026 FUTURE PROOF™

Manipulatives by the evidence

Held together, the studies compress into five rules a school can act on this term — none of which requires buying anything new.

Choose bland over branded. Perceptual richness is a cost paid in attention and transfer. Plain cubes beat googly-eyed counting bears for teaching; the charming objects belong to play, which has its own value and its own shelf (McNeil, Uttal, Jarvin & Sternberg, 2009).

Script the mathematics, not just the activity. The meta-analytic effects concentrate where objects are embedded in structured instruction with explicit links to notation — and thin out where children simply explore (Carbonneau, Marley & Selig, 2013). The question to ask of any blocks task: where, exactly, does the idea leave the blocks?

Plan the fade into the unit, in writing. Three representations per skill — concrete, schematic, symbolic — with planned exit points, is the sequence the comparative experiments reward (Fyfe, McNeil & Borjas, 2015). If the scheme of work never names the week the blocks go away, they won’t.

Assess in notation, not in wood. Object expertise can impersonate understanding — fluency with the material is not fluency with the maths (Uttal, Scudder & DeLoache, 1997). A child who can only show the answer with the apparatus is mid-journey, and the assessment should say so.

Keep the direction. Ground first, then strip — the reverse order, starting abstract and decorating later, shares every material and loses the benefit (Fyfe, McNeil, Son & Goldstone, 2014). Concrete is the on-ramp. The motorway is symbolic, because that is where the mathematics has to drive.