Math anxiety: the research on a learnable fear

Every teacher knows the child. Capable in every other lesson, fluent in reading, curious about the world — and visibly smaller the moment the math books come out. Hands that never go up. Homework that starts with tears. A test score that everyone, including the child, quietly accepts as the truth about ability.

Research has a different account of that child. Math anxiety — the tension and dread that arrives with numbers — is one of the best-measured emotions in education science. Its link to performance has been pooled across hundreds of studies. Its mechanism has been traced into working memory. Its transmission from adults to children has been caught in the act, at kitchen tables and in classrooms. And a set of interventions, some absurdly cheap, have been tested against it.

This article walks that evidence: the size of the link, the mechanism, the age it starts, the two transmission studies every school system should know, the global pattern — and what actually helps, as opposed to what adults usually do.

A fear with a measurement literature

Math anxiety has a formal definition — feelings of tension and apprehension that interfere with manipulating numbers and solving mathematical problems, in school and in ordinary life. The field has measured it with standardized scales since the early 1970s, long enough to accumulate an unusually deep evidence base (Hembree, 1990).

Two boundary facts matter before the numbers. First, math anxiety is not simply general anxiety wearing a costume: it correlates with test anxiety and general worry only moderately, and it attaches specifically to mathematical situations (Hembree, 1990). Second, it is not just a symptom of being bad at math. The correlation with ability is far too weak to explain the fear, and highly capable students report it too.

In other words: a real, specific, measurable condition — carried by a large share of every classroom, and by many of the adults standing in front of one. The first question a school leader should ask is how much it costs.

How big is the link, really?

The answer has been remarkably stable for thirty years. Hembree’s foundational meta-analysis — pooling hundreds of correlations — put the link between math anxiety and math performance at roughly r ≈ −.34 (Hembree, 1990). A second pooling a decade later landed at roughly −.27. And the modern meta-analysis, drawing on hundreds of samples from primary school to adulthood, settled at roughly r ≈ −.28 (Barroso et al., 2021).

A correlation near −.3 is moderate, not monstrous — worth taking seriously precisely because it is believable. It shows up at every grade level, somewhat stronger for teenagers than for young children, and it holds for boys and girls alike (Barroso et al., 2021).

One caution before anything else is built on it: this is a correlation, and the causal arrow runs both ways. Anxiety degrades performance, and poor performance breeds anxiety, in a loop that longitudinal studies suggest feeds itself (Barroso et al., 2021). The interesting question is where the loop can be cut. For that, the field had to find the mechanism.

The working-memory mechanism

Why would a feeling subtract points? The classic answer came from Ashcraft and Kirk. They measured working memory — the limited mental workspace that holds digits, carries, and intermediate steps — and found that highly math-anxious adults had smaller usable spans on computation-based tasks (Ashcraft & Kirk, 2001).

Then they made the mechanism visible. Participants did mental addition with carrying while holding letters in memory — a dual task that deliberately loads the workspace. As the load grew, everyone made more errors. But the highly anxious made far more, their error rates climbing steeply just when the task needed the workspace most (Ashcraft & Kirk, 2001).

The interpretation has held up across two decades: anxiety operates like a second task. Worry, self-monitoring, thoughts of escape — all of it runs in working memory, exactly where the math was supposed to run. The child is not refusing to think. The child is thinking about the fear, with the same limited machinery. That is why the deficit appears on multi-step problems and timed tests, and why it can vanish when stakes drop.

It starts in first grade

For years the field studied teenagers and adults, and schools could file the problem under adolescence. Then researchers asked six-year-olds. Ramirez and colleagues measured math anxiety in about 150 first and second graders — with pictures and simple language — and found it already present, already varying widely, and already correlated with achievement (Ramirez et al., 2013).

The detail that should reorganize early-years teaching: the negative link was concentrated in the children with the most working memory. Those children tend to use the most sophisticated strategies — decomposing numbers, holding intermediate steps — and those strategies are exactly what anxiety disrupts (Ramirez et al., 2013). The fear taxes the children with the most to lose. A classroom’s strongest young mathematicians can be its quietest casualties.

So by age six or seven the condition exists, varies, and costs. Where is it coming from? Children do not arrive hating numbers. Two studies caught the transmission directly.

Caught at home: the homework study

Maloney and colleagues followed roughly 440 first and second graders across a school year. At the start, they measured each parent’s math anxiety, and how often that parent helped with math homework. At both ends, they measured the children’s math achievement and math anxiety (Maloney et al., 2015).

The result is an interaction, and it is uncomfortable. When math-anxious parents helped with homework frequently, their children learned less math across the year and ended it more math-anxious. When the same kind of parent helped rarely, their children were fine. Parents low in math anxiety could help as often as they liked with no such cost. And the effect was specific to math — reading was untouched (Maloney et al., 2015).

What travels across the kitchen table is probably not arithmetic at all. It is sighs, tension, “I was never a math person,” frustration at a method that looks different from 1995. The study’s practical reading is precise: the problem is unstructured help delivered anxiously — which means the lever is structure, not distance.

Caught in class: teacher transmission

The classroom version of the same experiment of nature was published five years earlier. Beilock and colleagues followed 17 female first- and second-grade teachers and over a hundred of their students across a school year. At the start, the teachers’ own math anxiety predicted nothing about their students. By the end, it predicted the math achievement of the girls in their classrooms — and not the boys (Beilock et al., 2010).

The route was belief, not instruction. Girls with more math-anxious teachers were more likely to endorse the stereotype that boys are good at math and girls at reading, and the girls who absorbed that belief scored lower (Beilock et al., 2010). The teachers were not teaching worse arithmetic. They were modelling, to the children watching them for cues about their own kind, what a woman feels when the math books come out.

The study is small — seventeen classrooms in one district — and should be weighted accordingly. But its implication scales uncomfortably well, because elementary teaching is a heavily female profession, and elementary education majors report among the highest math-anxiety levels of any field of study (Beilock et al., 2010). A system that staffs its earliest math lessons with its most math-anxious adults has built a transmission mechanism and called it a workforce.

A global pattern

None of this is an American peculiarity. Using the PISA assessments — which survey fifteen-year-olds in dozens of countries about anxiety while testing their math — Foley and colleagues documented the same signature almost everywhere: within essentially every education system, students higher in math anxiety score lower (Foley et al., 2017).

The cross-country layer adds a twist. Higher-performing systems tend to report lower average math anxiety, yet within those same high-performing systems the individual anxiety-performance link is often strongest — competitive environments seem to sharpen the cost of the fear (Foley et al., 2017). For ministries, the message is blunt: raising standards without managing the emotional side channel can tighten the very link you want to break.

What helps

Here is the genuinely hopeful part: math anxiety responds to intervention better than most school problems, because the deficit it causes is partly artificial. The knowledge is often there. The workspace is occupied. Free the workspace and points come back.

The oldest evidence is the strongest on this point. In Hembree’s pooling, treatments aimed at the emotion itself — gradual desensitization and relaxation-based approaches, borrowed from clinical psychology — reduced anxiety and lifted the performance of treated students toward the levels of their low-anxious peers. Curriculum tweaks alone, without touching the emotion, did little (Hembree, 1990). The fear is the treatable part.

The cheapest evidence is more famous. Ramirez and Beilock had students spend ten minutes writing about their worries just before a high-pressure math test. In the lab, controls choked — accuracy dropped roughly 12% under pressure — while the writers actually improved by roughly 5%. In real ninth-grade final exams, the most anxious students who wrote performed on par with their calm peers, roughly a letter grade above anxious students who wrote about something else (Ramirez & Beilock, 2011). Writing the worry down seems to park it somewhere other than working memory.

The transmission studies carry their own instructions. For parents: the goal is structured help — a defined time, a calm tone, tools that carry the mathematical load — not less involvement (Maloney et al., 2015). For systems: teacher preparation that reduces teachers’ own math anxiety is child protection, not staff development (Beilock et al., 2010). And for daily classroom design, the mechanism is the guide: keep routine practice low-stakes and untimed where possible, because pressure is a working-memory load by another name (Ashcraft & Kirk, 2001). Adults sitting workplace assessments carry a parallel literature — we review it in our piece on test anxiety in adult assessment.

What the evidence doesn’t show

Math anxiety research is unusually convergent, which makes its honest edges easy to state — and worth stating.

• The core link is correlational. The ≈−.28 is not a causal estimate, and the arrow runs both ways: fear erodes performance and failure feeds fear (Barroso et al., 2021).
• Modest average, wide spread. A correlation near −.3 leaves most of the variance in achievement to everything else — anxiety is one dial on the board, not the board (Barroso et al., 2021).
• The teacher study is small. Seventeen classrooms, one district, girls only — a striking result that deserves replication weight, not law-of-nature weight (Beilock et al., 2010).
• Writing is a valve, not a cure. Expressive writing helps anxious students at test time; it teaches no mathematics and leaves the underlying fear in place (Ramirez & Beilock, 2011).
• No scaled school program exists. The treatments in the pooled evidence were mostly clinical-style, delivered to older students in small trials; a validated K-12 protocol at system scale has not been built (Hembree, 1990).
• The global data are self-report. PISA measures anxiety by questionnaire, and cultural response styles complicate country-to-country comparisons (Foley et al., 2017).

Where the evidence stops

1. 1The core link is correlational
2. 2Modest average, wide spread
3. 3The teacher study is small
4. 4Writing is a valve, not a cure
5. 5No scaled school program exists
6. 6The global data are self-report

© 2026 FUTURE PROOF™

Math anxiety by the evidence

Assembled, the literature reads as a short manual — one item each for teachers, parents, and the people who fund them.

Watch for it from first grade. The fear is measurable and already costly at six, and it hits hardest in the capable children using working-memory-heavy strategies (Ramirez et al., 2013). Avoidance — hands down, work skipped, guesses rushed — is a signal, not a personality.

Lower the stakes of daily practice. The deficit lives in working memory under pressure (Ashcraft & Kirk, 2001). Routine practice should be frequent, low-stakes and mostly untimed; save performance conditions for when they are actually being taught.

Re-equip anxious parents instead of benching them. The homework study’s lever is the how, not the whether: structured sessions, calm tone, tools that carry the mathematical load (Maloney et al., 2015). Telling parents to care less is neither realistic nor supported.

Treat teacher math confidence as a child outcome. The transmission route runs through the beliefs children absorb from anxious adults (Beilock et al., 2010). Elementary teacher preparation and ongoing support should budget for it explicitly.

Give anxious students a pressure valve. Ten minutes of expressive writing before high-stakes tests is one of the cheapest tested interventions in education (Ramirez & Beilock, 2011) — and treating the emotion directly has the oldest evidence of all (Hembree, 1990).