Preschoolers' Precision of the Approximate Number System Predicts Later School Mathematics Performance

For most of the twentieth century, the dominant assumption about mathematics learning was that it started in school. You sat down, someone taught you what numbers meant, and you built from there. The raw material was instruction. Then researchers started looking more carefully at infants — and at rats and at fish — and found something that complicated that picture entirely. There is a system in the human brain that performs mathematical operations long before anyone teaches it. It operates at birth, across species. A study by Michèle Mazzocco, Lisa Feigenson, and Justin Halberda showed that how sharp that system is at age four predicts how well a child will perform in school mathematics two years later — before a single formal lesson has been taught. The system is called the Approximate Number System, or ANS. The name is precise: it is approximate and it is a system. It does not deal in symbols or number words. It produces rough, nonverbal estimates of quantity — the kind that lets you glance at two piles of berries and know, instantly, that one is bigger without counting a single berry. This works across sensory modalities: visual arrays, sounds, even touch can trigger it. It is present in infants, in nonhuman animals, and in every human adult on earth. But the ANS has a fundamental limitation baked into its architecture. Its estimates are noisy, and the noise grows with the magnitude being represented. This is Weber's Law in action: the internal signal smears out as the number gets larger. An adult can easily distinguish one hundred dots from two hundred. Distinguishing one hundred from one hundred and ten is much harder — the ratio is tighter, and the noise overwhelms the difference. What improves across childhood is the precision of that signal. Young children can only reliably detect coarse ratios, like one-to-two. Older children handle finer ratios, like three-to-four. Adults do better still. But here is the critical fact: at any given age, there is wide variation between individuals. Some four-year-olds have a significantly sharper ANS than others. That variation is what Mazzocco and colleagues decided to track. The intellectual tension the study addresses is worth considering for a moment. Two competing hypotheses had been circulating in the field. One says the ANS is a foundation for formal mathematics — that the intuitive number sense you arrive at school with shapes how readily you absorb symbolic math. The other says it works the other way: learning formal mathematics sharpens the ANS, making it a byproduct of instruction rather than a precursor to it. These hypotheses are not mutually exclusive, but they have very different implications. The problem with prior evidence is that most studies linking the ANS and math performance were concurrent, meaning they measured both things at the same time in children already in school, or they were retrospective. In either case, instruction had already happened, and you could not untangle cause from effect. The fix is elegant in its logic. If you measure ANS precision before formal schooling begins and it still predicts later math achievement, the temporal ordering does real work. Instruction cannot have driven the ANS difference because instruction hadn't happened yet. That is exactly the design Mazzocco and colleagues used. They tested seventeen preschoolers — mean age four years and two months — on a non-symbolic dot discrimination task. Two arrays of familiar objects were flashed simultaneously, briefly enough that counting was impossible. The children simply had to judge which array was larger. Trials spanned a range of numerical ratios, from easy one-to-two comparisons down to tight eight-to-nine ones. Continuous visual cues like total surface area were carefully controlled by interleaving correlated and anti-correlated displays, so that the only reliable signal available was numerosity itself. Across children, performance ranged from forty-three to eighty-two percent correct — a substantial spread for kids the same age. The key measure of ANS precision is technically the Weber fraction, w — a psychophysical parameter describing how noisy the internal number signal is. Larger w means a noisier, less precise ANS. Because the number of trials per ratio was small, the authors used total percent correct, adjusted for age and display time, as a more stable proxy. Two years later, the same children — now around six years and eight months old — came back for follow-up testing. They were given the Test of Early Mathematics Ability, third edition, or TEMA-3, a standardized math assessment with strong reliability. They also took three subtests of the Wechsler Abbreviated Scale of Intelligence, the WASI, and a Rapid Automatized Naming task, the RAN, which measures how quickly a child can name sequences of letters, colors, and numbers — a proxy for language and reading-related abilities. Now for the results, which are what the whole design was built to deliver. Preschool ANS precision accounted for twenty-seven point eight percent of the variance in TEMA-3 scores two years later. The correlation was statistically significant: r-squared equals zero point twenty-eight, with a p-value of zero point zero three. When the four children who had performed at or near chance on the ANS task were excluded, leaving thirteen participants, that figure climbed to thirty-five point four percent of the variance, again significant. For comparison, concurrent full-scale IQ, measured at the follow-up visit, explained only about six point eight percent of the variance in TEMA-3 scores on its own — and was not statistically significant. The preschool number sense was, by that metric, a stronger predictor of school math than IQ measured at the same time as the math test. That alone is a striking finding. But the specificity result is what gives it real weight. ANS precision did not predict vocabulary — not even close, r-squared below zero point zero one, with a p-value of zero point eighty-nine. It did not predict Block Design or Matrix Reasoning performance. It did not predict rapid naming of colors or letters. The only hint of a relationship outside math was with rapid naming of numbers — RAN Numbers — where ANS precision accounted for thirty-two point four percent of the variance across the full sample. But that relationship disappeared when the analysis was restricted to the above-chance subgroup. So the picture that emerges is not one of a general cognitive advantage bleeding into math. Children with a sharper ANS at age four did not go on to be broadly smarter or quicker. They went on to be better at math specifically. Let that selectivity settle for a moment. The signal in the ANS task — a rough, nonverbal sense of approximate quantity, measured in preschoolers who couldn't read and hadn't started school — carried forward two years and landed precisely on mathematical performance, while leaving language, IQ, and perceptual reasoning untouched. That is not what you would expect from a general processing speed advantage or a general intelligence effect. It looks like something specific to numerical cognition. Mazzocco and colleagues are careful about what this does and does not prove. The sample is small: seventeen children, mostly white, all from middle socioeconomic backgrounds, with TEMA-3 scores clustered at the high end — standard scores ranged from ninety-eight to one hundred and thirty. That restricted range likely attenuated the correlations, meaning the true predictive relationship in a broader population could be even stronger. The study also used a single ANS measure at a single time point, and the authors are explicit that the mechanism remains unresolved. They are not claiming that ANS precision causes better math learning in a direct, proven, causal chain. They are claiming that the temporal ordering — ANS first, instruction second, math outcomes third — is consistent with the foundational hypothesis, and that prior work lacked that ordering. What the study opens up is a practical question. If ANS precision in preschool is a genuine predictor of school mathematics — and not merely a reflection of general smarts or early instruction — then wide variation in ANS acuity at age four matters in a way we hadn't fully reckoned with. It means some children arrive at kindergarten with a numerical foundation that is already stronger than others', through no fault of instruction or parenting. It suggests that identifying children with weaker ANS precision before school starts could create a window for early, targeted support. Mazzocco and colleagues call for larger, longitudinal studies to test that possibility, and to probe the mechanisms underneath. The finding here is the beginning of that argument, not its conclusion. But it is a beginning with a specific number attached to it: twenty-seven point eight percent of your school math performance at age six, predicted by how precisely you sensed approximate quantities at age four. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.