The Global Numeracy Crisis: How Many Children Lack Foundational Maths Skills?

Key Points

• UNICEF's October 2026 analysis of 56 household surveys across 46 countries found that pattern recognition is the weakest of the four foundational numeracy skills in 47 of those 56 surveys.

• Only 44.2% of children worldwide reached minimum proficiency in mathematics at the end of primary school in 2019, the most recent year with a published global figure, against 58.1% for reading.

• The World Bank's learning poverty indicator measures reading only, and no equivalent global headline measure of numeracy exists, so the scale of the maths crisis is systematically under-reported.

• In PISA 2025, published on 8 September 2026, the OECD average mathematics score fell to 463, the lowest recorded in the programme's history and down from 485 in 2015.

• Our analysis of UNICEF's own dataset found two distinct failure patterns that a single mastery rate hides: in 27 of 56 surveys most children who fail are one task short, while in 29 most pass no task at all.

Only 44.2% of children reach minimum proficiency in mathematics by the end of primary school, compared with 58.1% for reading. Reference year 2019, the most recent global aggregate published. UNESCO Institute for Statistics, SDG indicator 4.1.1, February 2026 release.

Ask how many children cannot do basic maths and you run into a strange problem. The world has a careful, well-funded, widely cited answer for reading. For maths it does not.

The World Bank's learning poverty measure, the number quoted in almost every discussion of the global learning crisis, counts children who cannot read and understand a simple story by age 10. There is no mathematics version. We checked the Learning Poverty Global Database directly: all 36 indicators in it belong to the reading family. The World Development Indicators catalogue returns no minimum-proficiency indicator for maths at all.

So the most famous statistic about the learning crisis is silent on half of it.

This article is an attempt to assemble the answer that does exist, from the surveys and assessments that do measure mathematics, and to be honest about where the measurements stop.

How we built it: we opened and verified more than 50 primary sources, meaning official reports, government statistics agencies, intergovernmental databases and peer-reviewed papers, taking every figure from the source document rather than from a summary of it. We then downloaded six official datasets and analysed them ourselves, including UNICEF's survey-level numeracy file, the TIMSS 2023 achievement exhibits and the OECD's PISA 2025 country results. Several findings below come from that analysis and have not been published elsewhere. Where a figure could not be verified at source, we say so rather than using it.

How Many Children Lack Foundational Maths Skills?

Worldwide, fewer than half of children reach minimum proficiency in mathematics by the end of primary school. The UNESCO Institute for Statistics puts the figure at 44.2% for reference year 2019. UNESCO's Global Education Monitoring Report 2026 states the same: 58% of students reach minimum proficiency in reading at the end of primary, and 44% in mathematics.

Regionally, the 2019 picture at the end of primary school looks like this.

Two things about this table deserve more attention than they usually get.

The first is the date. That is reference year 2019, and it is the most recent global aggregate published. The UIS release of February 2026 and the UN SDG Global Database agree on the value to five decimal places, and both stop at 2019. The GEM Report 2026 says plainly that the end-of-primary figure has not been updated since COVID-19. The best global number we have for children's maths predates the pandemic entirely.

The second is a trap. The same indicator at the end of lower secondary school shows a higher figure, 50.8%. Read quickly, that looks like children catching up. They are not.

These are shares of students still enrolled at that level, and in low and lower-middle-income countries roughly a quarter never reach the end of lower secondary. The children who left are the children who were struggling. The number rises because the sample improves, not because the children do.

What Foundational Numeracy Actually Means

Most people picture a maths test. The measurement that produces the clearest international picture of young children is narrower and stranger than that.

UNICEF's Foundational Learning Skills module, run inside the Multiple Indicator Cluster Surveys, tests one randomly selected child aged 7 to 14 in each sampled household. It happens at home, in the language the child chooses, whether or not the child attends school. That last detail matters enormously, and we will come back to it.

The numeracy test is 21 questions in four tasks.

The threshold is absolute. In UNICEF's words, a child has foundational numeracy skills "only if they answer all 21 correctly."

That is the standard being applied when you read that some percentage of children lack foundational numeracy. Not a pass mark of 50% or 70%. All 21.

On 19 October 2026, UNICEF published an analysis of 56 such surveys conducted between 2017 and 2024, covering 46 countries and territories. It is the richest picture of young children's numeracy currently available, and two of its three planned stories are still marked as forthcoming. We obtained the underlying dataset and analysed it ourselves. What follows includes UNICEF's published findings, which we reproduced as a check, and several patterns we found that UNICEF has not published.

The Four Skills, Ranked: What Children Fail Most

Across all 56 surveys, the four tasks are not equally hard. They fall into a consistent order.

UNICEF reports that pattern recognition is the weakest skill in 47 of the 56 surveys. We reproduced that count exactly from the dataset.

What UNICEF has not published, and what we found when we split the surveys into quartiles by overall performance, is that the order does not change with national wealth or school quality. It holds at the top and the bottom alike.

In the strongest quarter of surveys, nine children in ten can read a three-digit number, and roughly three in four can complete a number pattern. The gap narrows but never closes or reverses.

Within a single country the distance between the best and worst task runs to a median of 21.5 percentage points. In Tunisia's 2018 survey it reached 47.8 points, from number reading down to pattern recognition.

That consistency is the interesting part. A skill ordering that survives from Chad to Turkmenistan is unlikely to be an artefact of poverty, funding or teacher shortages. Something about the fourth task is harder, everywhere.

Pattern Recognition: The Skill Children Fail Most, and Research Understands Least

Here the data runs into the research literature, and the two do not line up.

Pattern recognition is the weakest foundational numeracy skill in the world. It also has the thinnest evidence base of the four.

The strongest study on early patterning, by Rittle-Johnson and colleagues in Child Development, followed 517 low-income American children from age 4 to 11. It found that patterning knowledge at the end of pre-kindergarten predicted fifth-grade mathematics achievement, with a standardised coefficient of .18, and that it remained a unique predictor when other skills were controlled. Shape knowledge, by contrast, never was.

But the same authors flagged a problem with their own measure. Its internal consistency ran between .56 and .63, which is low, and they wrote that their conclusions "remain tentative given the low reliability of the patterning measure in the current study."

A second study points the other way. Nguyen and colleagues, following 781 mostly low-income children, found patterning predicted fifth-grade achievement at .10, against .42 for counting and cardinality. In that sample, patterning was statistically indistinguishable from geometry and from measurement.

So the skill that children fail most often worldwide is the one where the research is least settled. And as Rittle-Johnson's team pointed out, United States content standards include shape but not patterning, a choice their findings "do not align with."

That curriculum point turns out to be sharper than it first appears. We checked the Common Core State Standards for Mathematics (CCSS) directly. Patterns appear in the Operations and Algebraic Thinking domain at three grades only: 3.OA.D.9, 4.OA.C.5 and 5.OA.B.3. There is no pattern standard in kindergarten, Grade 1 or Grade 2.

UNICEF's numeracy module is designed to test the skills expected by Grade 2 or 3. So a quarter of that test covers something the most widely used standards in the United States do not formally introduce until Grade 3. India's NCERT framework sequences early number work differently again.

We are not claiming patterning is the hidden key to mathematics, and we are not claiming the curriculum causes the failure. The honest position is narrower and more uncomfortable. A skill that a fifth of the world's foundational numeracy test is built on, and that children fail more than any other, sits on an evidence base thinner than the one under counting and appears late or not at all in early curricula. That is a gap in what we know, not a finding.

What the research does establish clearly is the primacy of symbolic number skills. A meta-analysis by Schneider and colleagues pooling 17,201 participants found that symbolic magnitude comparison, working with written numerals, correlated with mathematical competence at .302, more strongly than non-symbolic comparison at .241. A companion meta-analysis of 10,576 participants found number line estimation correlated at .443, rising to .491 after age nine. Nguyen's team found that when basic and advanced counting were entered together, advanced counting carried a coefficient of .33 while basic counting fell to .08 and lost significance.

The practical reading of that body of work: the skills that matter most involve numerals and number words, not an innate feel for quantity.

One caution belongs here, and the researchers state it themselves. Almost all of this evidence is correlational. Schneider's team wrote that such studies "leave unanswered whether numerical magnitude processing causally affects mathematical competence," and noted that causation may run backwards, since maths instruction has been shown to improve magnitude comparison. Predicts is not causes.

Two Crises Wearing One Label

This is the finding we think matters most, and it comes from asking a question the headline rate cannot answer: when a child fails, how badly do they fail?

Because mastery requires all 21 questions, a child who gets 20 right is counted exactly the same as a child who gets none. We looked at what is happening underneath.

Among children who fall short of full marks, across the 56 surveys:

• a median of 29.9% still pass three of the four tasks completely

• a median of 31.4% pass no task at all

• the median score among those who fall short is 14.4 out of 21

UNICEF reports that in 43 of 56 surveys, more than half of the children who fall short still answer every question correctly in at least one skill. We reproduced that figure.

Splitting the surveys by which failure pattern dominates produces two groups that look nothing alike.

These are two different problems that share a name. In Mongolia's 2018 survey, 60.4% of the children below full marks had already completed three of the four tasks. In other systems, most failing children have nothing to build on.

A country in the first group needs targeted teaching of one specific skill, most often pattern recognition. A country in the second needs foundational instruction from the beginning. A single national mastery rate tells a minister which number to be alarmed by and nothing about which of these two situations they are in.

The shape of the distribution makes the same point. A median of 43% of children score either 0 or 21, one extreme or the other. In seven surveys the two most common scores are exactly those two extremes: Afghanistan, Benin, Comoros, Guinea-Bissau, Malawi, Nigeria and Chad.

In Chad, 58% of children scored zero. In Pakistan's Balochistan province, 60.1% did.

Meanwhile, in Togo, the single most common score is 19 out of 21. Those children are two questions from being counted as numerate and are currently counted alongside children who answered nothing.

Country by Country: What the Regional Assessments Show

No single programme covers the world. Below is what each major regional and national assessment found, kept separate on purpose. The next section explains why combining them would be wrong.

South Asia

India runs two very different surveys, and the gap between them is instructive.

ASER 2024, a household survey reaching 649,491 children across 605 rural districts, tests arithmetic one child at a time, at home. Because it works from households rather than schools, it finds children who are out of school or absent.

The 2024 numbers are the highest in a decade for the younger grades, a genuine post-pandemic recovery. They also mean that two-thirds of children in Std III cannot do two-digit subtraction with borrowing, and that 5.5% cannot recognise single-digit numerals.

India's school-based national survey, PARAKH Rashtriya Sarvekshan 2024, tested 2,115,022 students in 74,229 schools on 4 December 2024. It reports scaled scores, not percentages, on a scale where 300 corresponds to answering half the test items correctly.

Grade 6 and Grade 9 both sit below the point representing half the items correct, and the Grade 9 figure is the lowest of any subject at any grade in the survey.

Sub-Saharan Africa

PASEC2019 assessed 14 francophone countries. At the end of primary school, 61.8% of pupils fell below the threshold PASEC defines as sufficient in mathematics. Reading was better at 52.1%.

PASEC also reports that 28.8% of pupils fall short at the start of primary school. The two figures sit side by side in the report's own conclusion. They come from separately calibrated scales, so the difference is not a measured decline, but the contrast is hard to ignore.

Latin America

UNESCO's ERCE 2019 study covered 16 countries and about 160,000 pupils. At Grade 6, 49.2% of children were at Nivel I, the lowest of four performance levels, in mathematics. The reading comparator was 23.3%. The lowest band in maths is more than twice as crowded as the lowest band in reading.

In the Dominican Republic, 77.0% of Grade 6 pupils were at Nivel I in mathematics.

Southeast Asia

SEA-PLM 2019 tested Grade 5 pupils in six countries against a band pegged to the end-of-primary SDG standard.

The spread between Viet Nam and Lao PDR, neighbours at similar income levels, is 84 percentage points on the same test in the same year.

The Rich World Is Not Exempt

PISA 2025, published on 8 September 2026 and covering 760,000 students in 91 countries and economies, recorded the worst mathematics performance in the programme's history. The OECD average fell to 463, from 472 in 2022 and 485 in 2015. The report states that mathematics "has fallen especially sharply after 2018."

We analysed the OECD's country file directly. Of the 74 countries and economies with a comparable 2022 figure, 60 recorded a lower mathematics score in 2025 and only 14 recorded a higher one, with a median change of about 9 points down.

The largest falls were in Malta, Slovenia, Israel, Serbia and Latvia. The largest rises were in Cambodia, Georgia, Jordan and the United Arab Emirates. These are point estimates; the OECD reports statistical significance separately.

In the United States, the 2024 NAEP results tell a two-speed story. Fourth-grade mathematics partially recovered from 2022 but remains below 2019. Eighth grade did not recover at all: the share scoring below NAEP Basic rose from 31.0% in 2019 to 39.4% in 2024. There is no 2026 NAEP mathematics release; those results are expected in 2027.

TIMSS 2023 found the same asymmetry internationally. Among 34 countries with comparable 2019 data at Grade 8, only three improved, 14 declined and 17 showed no change. At Grade 4 the picture was balanced, with 14 up and 13 down.

In England, 75% of Year 6 pupils met the expected standard in mathematics in summer 2026, up one point on 2025. A quarter did not.

What the Tests Cannot See

International assessments have a floor, and a growing number of systems are sitting on it.

TIMSS flags any country where more than 15% of students score too low for achievement to be estimated. At Grade 4 in 2023, two systems carried that flag. At Grade 8, five did. For Côte d'Ivoire, TIMSS states that average achievement was "not reliably measured because the percentage of students with achievement too low for estimation exceeds 25%."

The number of systems with a measurable floor problem more than doubles between Grade 4 and Grade 8. Where the crisis is deepest, the instrument stops describing it, and the published averages quietly understate the problem.

Why These Numbers Cannot Go on One Chart

It is tempting to line these percentages up. Doing so would be wrong, and the reasons are worth stating because they are rarely made explicit.

The thresholds measure different things. Each programme draws its line somewhere else, and some draw none at all.

• ASER: a single task performed correctly in front of an assessor.

• PASEC: Level 2 of a four-level scale.

• ERCE: Nivel I is the bottom band, a floor rather than a pass mark, so its polarity is reversed.

• SEA-PLM: a child is proficient in a band after answering half its questions.

• NAEP: Proficient is a deliberately aspirational American standard, not a grade-level expectation.

• England: a scaled score of 100 on a national test.

• SACMEQ: no threshold at all.

• PARAKH: scaled means, not proportions.

• UNICEF: all 21 questions correct.

The populations differ. This is the largest problem. ASER goes to households and so includes children who are out of school, absent or never enrolled. PASEC, SACMEQ, SEA-PLM, ERCE, NAEP and PISA test children who are in school on the day.

In systems with low enrolment or high absence, school-based results are flattered by exactly the children ASER reaches. England's figures are a near-census rather than a sample.

The years differ by more than a decade. SACMEQ's data is from 2013, while PASEC, ERCE and SEA-PLM are all from 2019, on the wrong side of the pandemic. NAEP, ASER and PARAKH are late 2024, and England is summer 2026.

Any cross-programme chart is also a time-travel chart.

The language differs. PASEC and SACMEQ test in the language of schooling, which for many children is not the language spoken at home. SEA-PLM's Philippine mathematics test was administered in English. Part of what a low score measures, in those systems, is comprehension of the test language.

Even the definition of maths differs. ASER tests arithmetic operations only. PASEC and SEA-PLM include measurement, geometry and data handling. NAEP covers five content areas including algebra.

What is safe is comparison within one programme over time, and between countries inside a single cycle of a single programme. That is how every table above is built.

One cross-check is worth reporting, because it tests whether these measurements agree at all. Five countries appear in both the UNICEF household surveys and PASEC's school-based assessment: Chad, DR Congo, Madagascar, Togo and Benin. Ranked from worst to best, the two programmes produce the same order except that Chad and DR Congo swap the bottom two places.

Two systems using different instruments, different populations and different years place these countries almost identically. With only five countries this is descriptive rather than statistical, but it is reassuring: the broad picture is not an artefact of one survey's method.

What Actually Works Against the Global Numeracy Crisis

The honest answer to "can this be fixed" is yes, by a measurable but modest amount, using methods that are well established.

The strongest causal evidence is the US Department of Education's What Works Clearinghouse practice guide on mathematics intervention in the elementary grades, published in 2021. All six of its recommendations carry a strong evidence rating, with the first resting on 43 studies.

The effect sizes are worth quoting accurately, because the eye-catching ones are the narrowest.

The number to quote for "what does intervention buy" is 0.31 on general mathematics achievement, not 1.47 on a narrow domain resting on ten studies. And one domain showed no significant effect at all, which is a useful corrective: systematic instruction does not improve everything equally.

On tutoring, a meta-analysis of 89 randomised controlled trials by Nickow, Oreopoulos and Quan found a pooled effect of 0.288 standard deviations, with mathematics at 0.268. Effects were larger when tutoring ran at least three days a week, when the tutor was a teacher or paraprofessional, in earlier grades, and during the school day. An earlier working-paper version reported 0.37 and is still widely quoted; the published figure is 0.288.

There is no credible source stating how many hours of instruction close a foundational numeracy gap. We looked for one. It does not exist.

Two further points from the research deserve to be stated carefully, because both are commonly overstated.

The first concerns whether early gaps widen. The intuition that they must is not settled. Morgan, Farkas and Wu found a Matthew effect in American data, with an initial-status to growth correlation of +.433. But Larsen and Little, analysing full state cohorts in Australia totalling 154,942 students across Grades 3 to 9, found the opposite: compensatory growth, with students below average at Grade 3 gaining more by Grade 9.

The defensible claim is that early gaps persist and strongly predict later standing, not that they automatically widen. Duncan's landmark study found school-entry maths predicted later achievement with an average coefficient of .33, against .13 for reading and .07 for attention, and the authors cautioned that their own coefficients might overstate causal effects.

The second concerns teaching. The National Research Council's Adding It Up describes five interwoven strands of mathematical proficiency and warns that "when students practice procedures they do not understand, there is a danger they will practice incorrect procedures," and that skills learned this way "are learned as isolated bits of knowledge." That is well established.

What is not established is the popular corollary that concepts must always be taught before procedures. Rittle-Johnson, Schneider and Star reviewed the evidence and found the relationship bidirectional, with alternative instructional orderings rarely compared.

What We Could Not Verify

Every figure in this article was read at its primary source. Several things we wanted were not available, and listing them is part of the method.

• There is no mathematics equivalent of learning poverty. We checked the Learning Poverty Global Database, the World Development Indicators catalogue and the World Bank's own briefs. Any "numeracy poverty" figure would have to be constructed, and we did not construct one.

• No global maths aggregate exists after 2019. Both UNESCO and the UN SDG database stop there.

• PISA 2025 publishes no combined below-Level-2 figure for mathematics. Volume I's focus domain was science. Only the gender split is published: 35% of girls and 32% of boys. We have not stated a combined figure because we could not read one at source.

• UNESCO's own metadata for indicator 4.1.1 contains an error. On page 5 it swaps reading and mathematics mid-sentence and cites a reading value that contradicts its own database. We did not use the figure in that sentence.

• UNICEF's data warehouse and its October 2026 analysis disagree on several countries, including Sierra Leone, where one scores a 19-question test out of 19 and the other appears to apply the 21-question rule. We used the October 2026 dataset throughout and say so.

• South Africa's SACMEQ IV gain of 57 points between 2007 and 2013 is unusually large and has been publicly questioned. We found no official adjudication, so we have not built on it. SACMEQ's own domain no longer resolves to the organisation.

• The SEA-PLM 2019 regional report is no longer online. Our figures come from the Philippines national report, which reproduces the six-country comparison.

• ERCE's full country-level table was behind a server that refused automated access. We used UNESCO's official country releases instead and report only the four countries we could read.

What to Remember About the Numeracy Crisis

• Fewer than half of the world's children reach minimum maths proficiency by the end of primary school, on a 2019 figure of 44.2%, and no official worldwide headline measure of numeracy exists at all.

• Pattern recognition is the weakest of the four foundational skills in 47 of 56 surveys, and that ordering holds in rich and poor countries alike.

• A single mastery rate conceals two different crises: in 27 of 56 surveys most failing children are one task short, while in 29 most pass no task at all.

• Mathematics is falling in the countries measured most recently, with 60 of 74 PISA participants scoring lower in 2025 than in 2022.

• Intervention works at roughly 0.3 standard deviations, which is real, modest, and nothing like a cure.

• For an individual child the useful question is not which percentile they sit in but which of the four skills is missing, and that is diagnosable in an afternoon: a free Bhanzu class starts by finding exactly that.

Research Behind This Article

1. UNICEF, Beyond the Full Score: A Closer Look at Children's Numeracy Skills (19 October 2026). data.unicef.org

2. UNESCO Institute for Statistics, SDG 4 Global and Thematic Indicators, February 2026 release, indicator 4.1.1. uis.unesco.org

3. UNESCO, Global Education Monitoring Report 2026: Monitoring Education in the SDGs (2026). unesco.org

4. OECD, PISA 2025 Results (Volume I): Future-Ready Students (September 2026), DOI 10.1787/73451bc5-en. oecd.org

5. IEA, TIMSS 2023 International Results in Mathematics and Science (4 December 2024). timss2023.org

6. ASER Centre, Annual Status of Education Report (Rural) 2024 (28 January 2025). asercentre.org

7. NCERT, PARAKH Rashtriya Sarvekshan 2024 (2025), ISBN 978-93-5729-675-5. parakh.ncert.gov.in

8. PASEC/CONFEMEN, PASEC2019: Qualité des systèmes éducatifs en Afrique subsaharienne francophone (2020). gpekix.org

9. National Center for Education Statistics, 2024 NAEP Mathematics Assessment (January 2025), NCES 2024217. nationsreportcard.gov

10. Institute of Education Sciences, Assisting Students Struggling with Mathematics: Intervention in the Elementary Grades, WWC 2021006 (March 2021). ies.ed.gov

11. National Research Council, Adding It Up: Helping Children Learn Mathematics (2001), National Academies Press. nationalacademies.org

12. Schneider, M. et al., "Associations of Number Line Estimation With Mathematical Competence: A Meta-analysis," Child Development 89(5), 1467-1484 (2018). doi.org

13. Duncan, G. J. et al., "School Readiness and Later Achievement," Developmental Psychology 43(6), 1428-1446 (2007). doi.org

14. World Bank, UNESCO, UNICEF et al., The State of Global Learning Poverty: 2022 Update (23 June 2022). worldbank.org