Early numeracy skills research: the first predictor.

Every school system runs on a quiet theory of what matters first. For decades that theory was letters and behaviour. Reception classrooms drilled the alphabet. Readiness checklists asked whether a child could sit on the carpet and take turns. Number got a corner of the day, and counting was treated as something children largely pick up on their own. The theory felt obvious. It was also, on the best available evidence, backwards.

The evidence in question is longitudinal — studies that measure children at school entry and then follow them for years. When researchers finally lined up the candidate readiness skills and let them compete for the same later outcomes, the ranking that emerged surprised nearly everyone. Early maths did not just hold its own against early reading. It beat it, roughly two to one. And attention and social behaviour, the staples of readiness talk, barely registered once the academic skills were in the model.

This article walks through that record. First the landmark six-dataset analysis and what it actually claims. Then the anatomy of early number sense, and the strange, bendy mental number line children start school with. Then the growth studies, the long-horizon forecasts that run all the way to algebra, and the intervention trials showing the base can be built deliberately. Last, the honest limits — because a predictor is not a destiny, and this literature is often oversold.

The study that reordered school readiness

In 2007, Greg Duncan and a large team published the analysis this whole field now orbits (Duncan et al., 2007). They took six longitudinal datasets from three countries — tens of thousands of children in total — and asked one disciplined question. Of the skills a child brings to the first day of school, which ones predict achievement years later, once you hold the others constant, along with family background and early cognitive ability?

The candidates were the usual suspects: early reading skills, early maths concepts, attention, and socioemotional skills such as getting on with peers. The outcome was achievement in later primary school, measured with standard tests. Because the six datasets differed in country, era and measures, any pattern that survived across all of them deserved to be taken seriously.

One pattern did. School-entry maths was the strongest predictor of later achievement in every dataset, with a pooled standardised coefficient of roughly 0.34. Early reading came in at roughly half that, around 0.17. Attention added a small but real slice, near 0.10. The socioemotional measures, once everything else was controlled, predicted later achievement at close to zero (Duncan et al., 2007).

Two details make the finding sharper. First, early maths predicted later reading about as well as early reading did — while the reverse was not true. Second, the pattern held for boys and girls, and for children from higher- and lower-income homes alike (Duncan et al., 2007). Whatever early number knowledge indexes, it is not a niche academic skill. It behaves like a load-bearing wall.

A word of caution belongs here, and the authors said it themselves. These are conditional associations, not experiments. No one randomly assigned children to know more about number. The coefficients say that early maths carries unique information about a child’s trajectory; they do not, by themselves, prove that raising early maths raises everything downstream. That question needs the intervention studies in section six. But as a map of where the readiness signal lives, the six-dataset result has replicated well and reset the field’s priorities.

What early number sense actually is

“Number sense” sounds like one thing. The research treats it as a small braid of skills that develop together in the preschool years. Counting is the visible strand: knowing the count words, saying them in order, and grasping that the last word tells you how many — the cardinal principle. Numeral knowledge links the spoken words to written digits. Magnitude comparison is the judgement strand: which is more, seven or nine? And underneath sits estimation — placing numbers, roughly, on an internal scale.

Part of that base appears to be far older than school. Humans, including infants, share with other animals an approximate number system — a built-in, rough sense of quantity that can tell 8 dots from 16 without counting. The system’s precision varies from person to person. In a well-known study, teenagers’ precision on this simple dot task correlated with their maths achievement all the way back to kindergarten records (Halberda, Mazzocco & Feigenson, 2008). The correlation is modest, and its direction is debated. But it suggests formal maths is threaded onto some deep perceptual hardware, not written on a blank slate.

What school actually builds on, though, is the symbolic braid: count words, digits, and comparisons between them. Those are the skills that carry the predictive weight in the school-readiness studies (Duncan et al., 2007). They are cheap to observe. A five-minute conversation — count these buttons, which pile has more, what number comes after six — samples most of the braid. Few signals in education are this predictive and this easy to collect.

The mental number line

The strand with the best story is estimation. Give a child a blank line with 0 at one end and 100 at the other. Ask them to mark where 18 goes. Adults space numbers evenly. Young children do something systematically different: they stretch the small numbers across most of the line and crush the large ones into the right-hand end. 18 lands near the middle. The pattern fits a compressed, roughly logarithmic mapping — equal jumps early on feel bigger than equal jumps later.

Robert Siegler and Julie Booth tracked how this internal line straightens out (Siegler & Booth, 2004). Kindergartners’ estimates on the 0–100 line followed the compressed curve. Second graders’ estimates were close to linear. The same bend then reappears on the harder 0–1,000 line and straightens again over the next few years. Growing up numerically is, in a real sense, learning that the number line is straight.

The straightening is not cosmetic. How linear a child’s estimates are correlates with their arithmetic and their memory for numbers, and the board-game experiments in section six suggest the link runs at least partly from representation to skill (Siegler & Booth, 2004). A child who feels that 18 sits near the middle of 0–100 is doing every sum on top of a warped map.

The growth that matters most

The six-dataset analysis measured children once, at school entry. Nancy Jordan and colleagues asked the sharper clinical question: which children should worry us, and when? They followed children from the start of kindergarten to the end of first grade, measuring “number competence” — counting, number knowledge, simple addition and subtraction — six times, then tested maths achievement in third grade (Jordan et al., 2009).

Both the starting point and the slope mattered. Where a child began in kindergarten predicted third-grade maths strongly, and how fast their number competence grew across kindergarten and first grade added predictive power on top (Jordan et al., 2009). Together, level and growth accounted for a large share of the variation in later achievement — before any third-grade teaching had happened.

The study also put an uncomfortable fact on the table. Children from low-income homes started kindergarten well behind on number competence, and the early gap, not later schooling, explained much of the income gap in third-grade maths (Jordan et al., 2009). The implication cuts both ways. Grim: disadvantage shows up in number knowledge before school gets a chance. Hopeful: the gap is specific, measurable at age five, and — as the intervention trials show — at least partly closable with cheap, targeted practice.

The long reach of early numbers

How far downstream does the early signal carry? Two studies mark the far end. Tyler Watts and colleagues followed children from age four and a half to age fifteen. Preschool number knowledge still predicted maths achievement at fifteen, and gains in maths from preschool to first grade predicted it on top of the starting point (Watts et al., 2014). A decade of schooling, friendships, teachers and luck sits between the two measurements. The signal survives it.

The second study picks up where early numeracy hands over. Siegler and colleagues took two national datasets — one American, one British — and asked which primary-school maths skills predict secondary-school algebra and overall maths achievement five to six years later. The winners, in both countries, were fractions and division. Their predictive weight held after controlling for IQ, reading, working memory, and family income and education (Siegler et al., 2012). Whole-number arithmetic is the floor; fractions are the hinge on which the door to algebra swings.

Read together, the chain is unusually complete for social science: preschool number sense forecasts school-entry competence; entry competence and its early growth forecast primary maths; primary fractions forecast secondary algebra (Siegler et al., 2012). Each link is a conditional association, not a law. But the same chain shows up in different countries, decades and datasets, and no rival readiness skill produces anything like it.

Can number sense be taught?

Correlation set the agenda; experiments answer the question. The most charming experiment in the literature used a board game. Ramani and Siegler worked with preschoolers from low-income families, whose number knowledge lagged their better-off peers on every measure. Half the children played a simple linear number game — ten numbered squares in a row, spin, move, and say the numbers as you go. The other half played the identical game with colours instead of numbers (Ramani & Siegler, 2008).

Total play time was about an hour: four sessions of fifteen to twenty minutes over two weeks. The number-game children improved on all four measures — number-line estimation, magnitude comparison, counting, and numeral identification — and the gains were still there nine weeks later. The colour-game children improved on none (Ramani & Siegler, 2008).

The design detail matters more than the charm. The game works, on the theory, because it is a number line a child can walk: square 7 is further along, takes more moves to reach, and takes longer to say. The linear spatial layout — the thing that maps distance onto magnitude — was the designed active ingredient, chosen to train the mental number line directly. The colour version, identical in every other way, was the control that isolated it (Ramani & Siegler, 2008).

Games are the small end of the intervention evidence. At the large end sit full preschool maths curricula. Clements and Sarama randomised classrooms serving mostly low-income children to their Building Blocks curriculum — a year-long programme organised around research-based learning trajectories for counting, comparison, shape and pattern — or to business as usual, with a third arm using a different maths curriculum. The Building Blocks classrooms ended the year roughly a full standard deviation ahead of business-as-usual on early maths, and roughly half a standard deviation ahead of the alternative curriculum (Clements & Sarama, 2008).

Effects that size are rare anywhere in education research. They say something specific: early number sense is not a fixed trait revealed at school entry. It is a curriculum-sensitive skill that ordinary preschools, given the right sequence, can build deliberately — and that an hour of well-designed play can measurably move.

What the evidence doesn’t show

The early-numeracy literature is strong, but it is routinely stretched past what it says. Six limits deserve to travel with the headline.

• Prediction is not causation. The readiness coefficients are conditional associations. They survive rich controls, but no study randomised early number knowledge, and third factors could drive part of every link in the chain (Duncan et al., 2007).
• Group signal, weak child-level screen. A coefficient of ≈0.34 is powerful for populations and modest for individuals. Many low scorers at five flourish; some high scorers struggle. Entry screens should trigger support, never labels or tracks (Jordan et al., 2009).
• Fade-out haunts the interventions. Preschool maths gains, like most preschool gains, shrink in the years after the programme ends — follow-ups of curriculum trials find the advantage attenuating as later grades fail to build on it (Clements & Sarama, 2008).
• The perceptual story is contested. The link between approximate-number precision and maths is small in meta-analyses, and its direction — hardware driving maths, or maths sharpening the hardware — remains unsettled (Halberda, Mazzocco & Feigenson, 2008).
• Measures vary by study. “Early maths” spans counting tests, applied-problem batteries and composite scores. The convergence across measures is reassuring, but exact coefficients are not interchangeable between studies (Watts et al., 2014).
• Long-horizon links thin out. The fractions-to-algebra and preschool-to-fifteen findings replicate in direction, but effect sizes shrink with distance, and cross-country coefficients differ in size (Siegler et al., 2012).

Where the evidence stops

1. 1Prediction is not causation
2. 2Group signal, weak child-level screen
3. 3Fade-out haunts the interventions
4. 4The perceptual story is contested
5. 5Measures vary by study
6. 6Long-horizon links thin out

© 2026 FUTURE PROOF™

Early numeracy by the evidence

Taken together, the literature reads as a short manual for schools, families and ministries — and it asks for surprisingly little money.

Measure number sense at entry, and measure its growth. A brief screen of counting, comparison and numeral knowledge at school entry, repeated across the first two years, captures the level-plus-slope signal that best forecasts trouble (Jordan et al., 2009). The point is triage: children flagged early get practice early, while the skills are cheap to build.

Give number the same daily status as letters. The readiness evidence says early maths carries at least as much forward weight as early reading — including for reading (Duncan et al., 2007). A preschool day with twenty minutes of stories and zero minutes of number talk is running the old theory, not the evidence.

Use linear, spatial representations. Number lines on the wall and on the floor, board games with numbered squares in a row, counting with distance and movement — these target the representation that predicts and responds (Ramani & Siegler, 2008). An hour of the right game is not a metaphor; it is a measured dose.

Buy sequence, not worksheets. The big curricular effects came from learning trajectories — teaching counting, comparison and pattern in a researched order at the right grain (Clements & Sarama, 2008). And plan the follow-through: fade-out is what happens when year two ignores what year one built.

Protect fractions later. The early-years signal hands over to fractions and division around ages nine to twelve — the strongest known lever for algebra readiness (Siegler et al., 2012). Early numeracy is the first predictor, not the last; systems need the whole chain.