education poverty

About . Click to expand section.

• Our History
• Team & Board
• Transparency and Accountability

What We Do . Click to expand section.

• Cycle of Poverty
• Climate & Environment
• Emergencies & Refugees
• Health & Nutrition
• Livelihoods
• Gender Equality
• Where We Work

Take Action . Click to expand section.

• Attend an Event
• Partner With Us
• Fundraise for Concern
• Work With Us
• Leadership Giving
• Humanitarian Training
• Newsletter Sign-Up

Donate . Click to expand section.

• Give Monthly
• Donate in Honor or Memory
• Leave a Legacy
• DAFs, IRAs, Trusts, & Stocks
• Employee Giving

How does education affect poverty?

For starters, it can help end it.

Aug 10, 2023

Access to high-quality primary education and supporting child well-being is a globally-recognized solution to the cycle of poverty. This is, in part, because it also addresses many of the other issues that keep communities vulnerable.

Education is often referred to as the great equalizer: It can open the door to jobs, resources, and skills that help a person not only survive, but thrive. In fact, according to UNESCO, if all students in low-income countries had just basic reading skills (nothing else), an estimated 171 million people could escape extreme poverty. If all adults completed secondary education, we could cut the global poverty rate by more than half. 

At its core, a quality education supports a child’s developing social, emotional, cognitive, and communication skills. Children who attend school also gain knowledge and skills, often at a higher level than those who aren’t in the classroom. They can then use these skills to earn higher incomes and build successful lives.

Here’s more on seven of the key ways that education affects poverty.

Go to the head of the class

Get more information on Concern's education programs — and the other ways we're ending poverty — delivered to your inbox.

1. Education is linked to economic growth

Education is the best way out of poverty in part because it is strongly linked to economic growth. A 2021 study co-published by Stanford University and Munich’s Ludwig Maximilian University shows us that, between 1960 and 2000, 75% of the growth in gross domestic product around the world was linked to increased math and science skills. 

“The relationship between…the knowledge capital of a nation, and the long-run [economic] rowth rate is extraordinarily strong,” the study’s authors conclude. This is just one of the most recent studies linking education and economic growth that have been published since 1990.

“The relationship between…the knowledge capital of a nation, and the long-run [economic] growth rate is extraordinarily strong.” — Education and Economic Growth (2021 study by Stanford University and the University of Munich)

2. Universal education can fight inequality

A 2019 Oxfam report says it best: “Good-quality education can be liberating for individuals, and it can act as a leveler and equalizer within society.” 

Poverty thrives in part on inequality. All types of systemic barriers (including physical ability, religion, race, and caste) serve as compound interest against a marginalization that already accrues most for those living in extreme poverty. Education is a basic human right for all, and — when tailored to the unique needs of marginalized communities — can be used as a lever against some of the systemic barriers that keep certain groups of people furthest behind. 

For example, one of the biggest inequalities that fuels the cycle of poverty is gender. When gender inequality in the classroom is addressed, this has a ripple effect on the way women are treated in their communities. We saw this at work in Afghanistan , where Concern developed a Community-Based Education program that allowed students in rural areas to attend classes closer to home, which is especially helpful for girls.

Four ways that girls’ education can change the world

Gender discrimination is one of the many barriers to education around the world. That’s a situation we need to change.

3. Education is linked to lower maternal and infant mortality rates

Speaking of women, education also means healthier mothers and children. Examining 15 countries in sub-Saharan Africa, researchers from the World Bank and International Center for Research on Women found that educated women tend to have fewer children and have them later in life. This generally leads to better outcomes for both the mother and her kids, with safer pregnancies and healthier newborns. 

A 2017 report shows that the country’s maternal mortality rate had declined by more than 70% in the last 25 years, approximately the same amount of time that an amendment to compulsory schooling laws took place in 1993. Ensuring that girls had more education reduced the likelihood of maternal health complications, in some cases by as much as 29%. 

4. Education also lowers stunting rates

Children also benefit from more educated mothers. Several reports have linked education to lowered stunting , one of the side effects of malnutrition. Preventing stunting in childhood can limit the risks of many developmental issues for children whose height — and potential — are cut short by not having enough nutrients in their first few years.

In Bangladesh , one study showed a 50.7% prevalence for stunting among families. However, greater maternal education rates led to a 4.6% decrease in the odds of stunting; greater paternal education reduced those rates by 2.9%-5.4%.  A similar study in Nairobi, Kenya confirmed this relationship: Children born to mothers with some secondary education are 29% less likely to be stunted.

What is stunting?

Stunting is a form of impaired growth and development due to malnutrition that threatens almost 25% of children around the world.

5. Education reduces vulnerability to HIV and AIDS…

In 2008, researchers from Harvard University, Imperial College London, and the World Bank wrote : “There is a growing body of evidence that keeping girls in school reduces their risk of contracting HIV. The relationship between educational attainment and HIV has changed over time, with educational attainment now more likely to be associated with a lower risk of HIV infection than earlier in the epidemic.” 

Since then, that correlation has only grown stronger. The right programs in schools not only reduce the likelihood of young people contracting HIV or AIDS, but also reduce the stigmas held against people living with HIV and AIDS.

6. …and vulnerability to natural disasters and climate change

As the number of extreme weather events increases due to climate change, education plays a critical role in reducing vulnerability and risk to these events. A 2014 issue of the journal Ecology and Society states: “It is found that highly educated individuals are better aware of the earthquake risk … and are more likely to undertake disaster preparedness.… High risk awareness associated with education thus could contribute to vulnerability reduction behaviors.”

The authors of the article went on to add that educated people living through a natural disaster often have more of a financial safety net to offset losses, access to more sources of information to prepare for a disaster, and have a wider social network for mutual support.

Climate change is one of the biggest threats to education — and growing

Last August, UNICEF reported that half of the world’s 2.2 billion children are at “extremely high risk” for climate change, including its impact on education. Here’s why.

7. Education reduces violence at home and in communities

The same World Bank and ICRW report that showed the connection between education and maternal health also reveals that each additional year of secondary education reduced the chances of child marriage — defined as being married before the age of 18. Because educated women tend to marry later and have fewer children later in life, they’re also less likely to suffer gender-based violence , especially from their intimate partner. 

Girls who receive a full education are also more likely to understand the harmful aspects of traditional practices like FGM , as well as their rights and how to stand up for them, at home and within their community.

Fighting FGM in Kenya: A daughter's bravery and a mother's love

Marsabit is one of those areas of northern Kenya where FGM has been the rule rather than the exception. But 12-year-old student Boti Ali had other plans.

Education for all: Concern’s approach

Concern’s work is grounded in the belief that all children have a right to a quality education. Last year, our work to promote education for all reached over 676,000 children. Over half of those students were female. 

We integrate our education programs into both our development and emergency work to give children living in extreme poverty more opportunities in life and supporting their overall well-being. Concern has brought quality education to villages that are off the grid, engaged local community leaders to find solutions to keep girls in school, and provided mentorship and training for teachers.

More on how education affects poverty

6 Benefits of literacy in the fight against poverty

Child marriage and education: The blackboard wins over the bridal altar

Project Profile

Right to Learn

Sign up for our newsletter.

Get emails with stories from around the world.

You can change your preferences at any time. By subscribing, you agree to the Terms of Use and Privacy Policy.

• Get involved

The transformative power of education in the fight against poverty

October 16, 2023.

Zubair Junjunia, a Generation17 young leader and the Founder of ZNotes, presents at EdTechX.

Zubair Junjunia

Generation17 Young Leader and founder of ZNotes

Time and again, research has proven the incredible power of education to break poverty cycles and economically empower individuals from the most marginalized communities with dignified work and upward social mobility. 

Research at UNESCO has shown that world poverty would be more than halved if all adults completed secondary school. And if all students in low-income countries had just basic reading skills, almost 171 million people could escape extreme poverty. 

With such irrefutable evidence, how do we continue to see education underfunded globally? Funding for education as a share of national income has not changed significantly over the last decade for any developing country. And to exacerbate that, the COVID-19 shock pushed the level of learning poverty to an estimated 70 percent .

I have devoted the past decade of my life to fighting educational inequality, a journey that began during my school years. This commitment led to the creation of ZNotes , an educational platform developed for students, by students. ZNotes was born out of the problem I witnessed first-hand; the inequities in end-of-school examination, which significantly influence access to higher education and career opportunities. It is designed as a platform where students can share their notes and access top-quality educational materials without any limitations. ZNotes fosters collaborative learning through student-created content within a global community and levels the academic playing field with a student-empowered and technology-enabled approach to content creation and peer learning. 

Although I started ZNotes as a solo project, today, it has touched the lives of over 4.5 million students worldwide, receiving an impressive 32 million hits from students across more than 190 countries, especially serving students from emerging economies. We’re proud to say that today, more than 90 percent of students find ZNotes resources useful and feel more confident entering exams , regardless of their socio-economic background. These globally recognized qualifications empower our learners to access tertiary education and enter the world of work.

Sixteen-year-old Zubair set up a blog to share the resources he created for his IGCSE exams. Through word of mouth, his revision notes were discovered by students all over the world and ZNotes was born.

In rapidly changing job market, young people must cultivate resilience and adaptability. World Economic Forum highlights the importance of future skills, encompassing technical, cognitive, and interpersonal abilities. Unfortunately, many educational systems, especially in under-resourced regions, fall short in equipping youth with these vital skills.

To address this challenge, I see innovative technology as a crucial tool both within and beyond traditional school systems. As the digital divide narrows and access to devices and internet connectivity becomes more affordable, delivering quality education and personalized support is increasingly achievable through technology. At ZNotes, we are reshaping the role of students, transforming them from passive consumers to active creators and proponents of education. Empowering youth through a community-driven approach, students engage in peer learning and generate quality resources on an online platform.

Participation in a global learning community enhances young people's communication and collaboration skills. ZNotes fosters a sense of global citizenship, enabling learners to communicate with a diverse range of individuals across race, gender, and religion. Such spaces also result in redistributing social capital as students share advice for future university, internship and career pathways.

“Studying for 14 IGCSE subjects wasn't easy, but ZNotes helped me provide excellent and relevant revision material for all of them. I ended up with 7 A* 7 A, and ZNotes played a huge role. I am off to Cornell University this fall now. A big thank you to the ZNotes team!"

Alongside ensuring our beneficiaries are equipped with the resources and support they need to be at a level playing field for such high stakes exams, we also consider the skills that will set them up for success in life beyond academics. Especially for the hundreds of young people who join our internship and contribution programs , they become part of a global social impact startup and develop both academic skills and also employability skills. After engaging with our internship programs, 77% of interns reported improved candidacy for new jobs and internships. 

ZNotes addresses the uneven playing field of standardized testing with a student-empowered and technology-enabled approach for content creation and peer learning.

A few years ago, Jess joined our team as a Social Impact Analyst intern having just completed her university degree while she continued to search for a full-time role. She was able to apply her data analytics skills from a theoretical degree into a real-world scenario and was empowered to play an instrumental role in understanding and developing a Theory of Change model for ZNotes. In just 6 months, she had been able to develop the skills and gain experiences that strengthened her profile. At the end of internship, she was offered a full-time role at a major news and media agency that she is continuing to grow in!

Jess’s example applies to almost every one of our interns . As another one of them, Alexa, said “ZNotes offers the rare and wonderful opportunity to be at the center of meaningful change”.

Being part of an organization making a significant impact is profoundly inspiring and empowering for young people, and assuming high-responsibility roles within such organizations accelerates their skills development and sets them apart in the eyes of prospective employers.

On the International Day for the Eradication of Poverty, it is a critical moment to reflect and enact on the opportunity that we have to achieving two key SDGs, Goal 1 and 4, by effectively funding and enabling access to quality education globally.

You are using an outdated browser. Please upgrade your browser or activate Google Chrome Frame to improve your experience.

Thanks for signing up as a global citizen. In order to create your account we need you to provide your email address. You can check out our Privacy Policy to see how we safeguard and use the information you provide us with. If your Facebook account does not have an attached e-mail address, you'll need to add that before you can sign up.

This account has been deactivated.

Please contact us at [email protected] if you would like to re-activate your account.

Lack of access to education is a major predictor of passing poverty from one generation to the next, and receiving an education is one of the top ways to achieve financial stability.

In other words: education and poverty are directly linked.

Increasing access to education can equalize communities, improve the overall health and longevity of a society , and help save the planet .

The problem is that about 258 million children and youth are out of school around the world, according to UNESCO data released in 2018. 

Children do not attend school for many reasons — but they all stem from poverty.

Here are all the statistics, facts, and answers to questions you might have that shed light on the connection between poverty and education. 

How does poverty affect education?

Families living in poverty often have to choose between sending their child to school or providing other basic needs. Even if families do not have to pay tuition fees, school comes with the added costs of uniforms, books, supplies, and/or exam fees. 

Countries across sub-Saharan Africa, where the world’s poorest children live, have made a concerted effort to abolish school fees . While the ratio of students completing lower secondary school increased  in the region from 23% in 1990 to 42% in 2014, enrollment is low compared to the 75% global ratio. School remains too expensive for the poorest families. Some children are forced to stay at home doing chores or need to work. In other places, especially in crisis and conflict areas with destroyed infrastructure and limited resources, unaffordable private schools are sometimes the only option .

Why does poverty stop girls from going to school? 

Poverty is the most important factor that determines whether or not a girl can access education, according to the World Bank. If families cannot afford the costs of school, they are more likely to send boys than girls. Around 15 million girls will never get the chance to attend school, compared to 10 million boys. 

Read More: These Are the Top 10 Best and Worst Countries for Education in 2016 

Gender inequality is more prevalent in low-income countries. Women often perform more unpaid work, have fewer assets, are exposed to gender-based violence, and are more likely to be forced into early marriage, all limiting their ability to fully participate in society and benefit from economic growth. 

When girls face barriers to education early on, it is difficult for them to recover. Child marriage is one of the most common reasons a girl might stop going to school. More than 650 million women globally have already married under the age of 18. For families experiencing financial hardship, child marriage reduces their economic burden , but it ends up being more difficult for girls to gain financial independence if they are unable to access a quality education.

Lack of access to adequate menstrual hygiene management also stops many girls from attending school. Some girls cannot afford to buy sanitary products or they do not have access to clean water and sanitation to clean themselves and prevent disease. If safety is a concern due to lack of separate bathrooms, girls will stay home from school to avoid putting themselves at risk of sexual assault or harassment. 

Read More: 10 Barriers to Education Around the World

An educated girl is not only likely to increase her personal earning potential but can help reduce poverty in her community, too. 

“Educated girls have fewer, healthier, and better-educated children,” according to the Global Partnership for Education.

When countries invest in girls’ education, it sees an increase in female leaders, lower levels of population growth, and a reduction of contributions to climate change. 

Can education help break the cycle of poverty? 

Education promotes economic growth because it provides skills that increase employment opportunities and income. Nearly 60 million people could escape poverty if all adults had just two more years of schooling, and 420 million people could be lifted out of poverty if all adults completed secondary education, according to UNESCO. 

Education increases earnings by roughly 10% per each additional year of schooling. For each $1 invested in an additional year of schooling, earnings increase by $5 in low-income countries and $2.5 in lower-middle income countries. 

Read More: 264 Million Children Are Denied Access To Education, New Report Says

Education reduces many issues that stop people from living healthy lives, including infant and maternal deaths, stunting, infant and maternal deaths, vulnerability to HIV/AIDS, and violence.

How can we end extreme poverty through education?

There are more children enrolled in school than ever before — developing countries reached a 91% enrollment rate in 2015 — but we must fully close the gap. 

World leaders gathered at the United Nations headquarters to address the disparity in 2015 and set 17 Global Goals to end extreme poverty by 2030. Global Goal 4: Quality Education aims to "end poverty in all its forms everywhere."

Read More: How We Can Be the Generation to End Extreme Poverty

The first step to achieving quality education for all is acknowledging that it is a vital part of sustainable development. Citizens, governments, corporations, and philanthropists all have an important role to play. Learn how to ensure global access to education to end poverty by taking action here .

Global Citizen Explains

Defeat Poverty

Understanding How Poverty is the Main Barrier to Education

Feb. 7, 2020

• High contrast
• Press Centre

Search UNICEF

Every child has the right to learn..

• Available in:

Overview  |  What we do  |  Reports  |  Data  |  News

A child’s right to education entails the right to learn. Yet, for too many children across the globe, schooling does not lead to learning.

Over 600 million children worldwide are unable to attain minimum proficiency levels in reading and mathematics, even though two thirds of them are in school. For out-of-school children, foundational skills in literacy and numeracy are further from grasp.

Children are deprived of education for various reasons. Poverty remains one of the most obstinate barriers. Children living through economic fragility, political instability, conflict or natural disaster are more likely to be cut off from schooling – as are those with disabilities, or from ethnic minorities. In some countries, education opportunities for girls remain severely limited.

Even in schools, a lack of trained teachers, inadequate education materials and poor infrastructure make learning difficult for many students. Others come to class too hungry, ill or exhausted from work or household tasks to benefit from their lessons.

Compounding these inequities is a digital divide of growing concern: Most of the world’s school-aged children do not have internet connection in their homes, restricting their opportunities to further their learning and skills development.

Without quality education, children face considerable barriers to employment later in life. They are more likely to suffer adverse health outcomes and less likely to participate in decisions that affect them – threatening their ability to shape a better future for themselves and their societies.

Education is a basic human right. In 147 countries around the world, UNICEF works to provide quality learning opportunities that prepare children and adolescents with the knowledge and skills they need to thrive. We focus on:

Equitable access : Access to quality education and skills development must be equitable and inclusive for all children and adolescents, regardless of who they are or where they live. We make targeted efforts to reach children who are excluded from education and learning on the basis of gender, disability, poverty, ethnicity and language. 

Quality learning : Outcomes must be at the centre of our work to close the gap between what students are learning and what they need to thrive in their communities and future jobs. Quality learning requires a safe, friendly environment, qualified and motivated teachers, and instruction in languages students can understand. It also requires that education outcomes be monitored and feed back into instruction.

Education in emergencies : Children living through conflict, natural disaster and displacement are in urgent need of educational support. Crises not only halt children’s learning but also roll back their gains. In many emergencies, UNICEF is the largest provider of educational support throughout humanitarian response, working with UNHCR, WFP and other partners.

Our programmes

Our strategy

Get involved

Help us tackle the learning crisis.

Global Annual Results Report 2023: Humanitarian action

Progress, results achieved and lessons from 2023 in UNICEF humanitarian action

Global Report on Early Childhood Care and Education

The right to a strong foundation

Building Strong Foundations

Brighter futures through education for health and well-being

Skills for a green transition

Solutions for youth on the move

Data and insights

Our research

Our insights

Ahead of Day of the African Child UNICEF says African governments not spending what they need to secure quality education for continent’s children

1,000 days of education – equivalent to three billion learning hours – lost for Afghan girls

Ukraine’s recovery is dependent on the recovery of children’s education

Children call for access to quality climate education

On Earth Day, UNICEF urges governments to empower every child with learning opportunities to be a champion for the planet

• Ending poverty
• Facts about poverty
• Education & poverty

Global poverty and education

Facts & stats about world poverty and education.

• Global poverty facts
• Health & poverty
• Empowerment & poverty
• Employment & poverty
• Giving & poverty
• Technology & poverty
• Hunger & poverty
• Poverty in the U.S.
• Poverty in Africa

The research is conclusive: When we reduce barriers to education, we set children up to thrive. It’s not only about knowledge and numbers, access to education reduces a child’s involvement in gangs and drugs, and lowers the amount of teen pregnancies. Education leads to healthier childhoods and, ultimately, to greater economic prospects as adults. Your sponsorship or gift helps provide children access to life-changing education programs in the communities we serve, as well as crucial health and dental services,  life-skills and career-placement workshops, and more.

Sponsor a child       Make a gift Learn about our Education Programs

Children who participate in early childhood development achieve higher education levels and make more money as adults.

In developing nations, 60% of 10 ten-year-olds suffer from learning poverty, unable to read/understand simple stories.

For every year a girl remains in school, her average lifetime income increases 10% and odds of an early marriage shrink.

A child’s survival beyond age 5 increases by 31% when a mother has a high school degree, compared to no education.

Despite the shrinking gender gap, 8% fewer girls complete lower secondary school than boys. 

The share of youth not employed or participating in education or job training rose to 23.3%, the highest increase in 15 years.

Global studies find there is a 9% increase in hourly earnings for every extra year of schooling a child receives. 

During COVID, children in Latin America and the Caribbean lost an average of 225 full days of school.

Access to education

We believe providing children access to education and resources ushers them into a world of ideas and knowledge, and creates lasting change in their lives. Supporters provide access to tutoring, computer courses, libraries and more, setting children up for a lifetime of success.

• WHO Youth Violence Research, 2009
• UNICEF The State of the World’s Children, 2009
• UNESCO, 2012
• World Bank eLibrary. “Returns on Investments in Education,” 2002
• UNESCO Global Education Monitoring Report, 2017
• The World Bank  2015
• The World Bank  2017
• Globalpartnership.org, Education Data Highlights
• Un.org, Education
• Results.Org, World Poverty and What you Can Do About It
• Lifewater.org, 9 Poverty Statistics that Everyone Should Know, January 28, 2020

This site uses cookies to improve your experience. You can restrict cookies through your browser; however that may impair site functionality.

Are you sure?

My order summary

Informing and Advancing Effective Policy

• Newsletters
• Education Gap Grows Between Rich and Poor, Studies Say

By SABRINA TAVERNISE

Education was historically considered a great equalizer in American society, capable of lifting less advantaged children and improving their chances for success as adults. But a body of recently published scholarship suggests that the achievement gap between rich and poor children is widening, a development that threatens to dilute education’s leveling effects.

It is a well-known fact that children from affluent families tend to do better in school. Yet the income divide has received far less attention from policy makers and government officials than gaps in student accomplishment by race.

Now, in analyses of long-term data published in recent months, researchers are finding that while the achievement gap between white and black students has narrowed significantly over the past few decades, the gap between rich and poor students has grown substantially during the same period.

“We have moved from a society in the 1950s and 1960s, in which race was more consequential than family income, to one today in which family income appears more determinative of educational success than race,” said Sean F. Reardon, a Stanford University sociologist. Professor Reardon is the author of a study that found that the gap in standardized test scores between affluent and low-income students had grown by about 40 percent since the 1960s, and is now double the testing gap between blacks and whites.

In another study, by researchers from the University of Michigan , the imbalance between rich and poor children in college completion — the single most important predictor of success in the work force — has grown by about 50 percent since the late 1980s.

The changes are tectonic, a result of social and economic processes unfolding over many decades. The data from most of these studies end in 2007 and 2008, before the recession’s full impact was felt. Researchers said that based on experiences during past recessions, the recent downturn was likely to have aggravated the trend.

“With income declines more severe in the lower brackets, there’s a good chance the recession may have widened the gap,” Professor Reardon said. In the study he led, researchers analyzed 12 sets of standardized test scores starting in 1960 and ending in 2007. He compared children from families in the 90th percentile of income — the equivalent of around $160,000 in 2008, when the study was conducted — and children from the 10th percentile, $17,500 in 2008. By the end of that period, the achievement gap by income had grown by 40 percent, he said, while the gap between white and black students, regardless of income, had shrunk substantially.

The connection between income inequality among parents and the social mobility of their children has been a focus of President Obama as well as some of the Republican presidential candidates.

One reason for the growing gap in achievement, researchers say, could be that wealthy parents invest more time and money than ever before in their children (in weekend sports, ballet, music lessons, math tutors, and in overall involvement in their children’s schools), while lower-income families, which are now more likely than ever to be headed by a single parent, are increasingly stretched for time and resources. This has been particularly true as more parents try to position their children for college, which has become ever more essential for success in today’s economy.

A study by Sabino Kornrich, a researcher at the Center for Advanced Studies at the Juan March Institute in Madrid, and Frank F. Furstenberg, scheduled to appear in the journal Demography this year, found that in 1972, Americans at the upper end of the income spectrum were spending five times as much per child as low-income families. By 2007 that gap had grown to nine to one; spending by upper-income families more than doubled, while spending by low-income families grew by 20 percent.

“The pattern of privileged families today is intensive cultivation,” said Dr. Furstenberg, a professor of sociology at the University of Pennsylvania.

The gap is also growing in college. The University of Michigan study, by Susan M. Dynarski and Martha J. Bailey, looked at two generations of students, those born from 1961 to 1964 and those born from 1979 to 1982. By 1989, about one-third of the high-income students in the first generation had finished college; by 2007, more than half of the second generation had done so. By contrast, only 9 percent of the low-income students in the second generation had completed college by 2007, up only slightly from a 5 percent college completion rate by the first generation in 1989.

James J. Heckman, an economist at the University of Chicago, argues that parenting matters as much as, if not more than, income in forming a child’s cognitive ability and personality, particularly in the years before children start school.

“Early life conditions and how children are stimulated play a very important role,” he said. “The danger is we will revert back to the mindset of the war on poverty, when poverty was just a matter of income, and giving families more would improve the prospects of their children. If people conclude that, it’s a mistake.”

Meredith Phillips, an associate professor of public policy and sociology at the University of California, Los Angeles, used survey data to show that affluent children spend 1,300 more hours than low-income children before age 6 in places other than their homes, their day care centers, or schools (anywhere from museums to shopping malls). By the time high-income children start school, they have spent about 400 hours more than poor children in literacy activities, she found.

Charles Murray, a scholar at the American Enterprise Institute whose book, “Coming Apart: The State of White America, 1960-2010,” was published Jan. 31, described income inequality as “more of a symptom than a cause.”

The growing gap between the better educated and the less educated, he argued, has formed a kind of cultural divide that has its roots in natural social forces, like the tendency of educated people to marry other educated people, as well as in the social policies of the 1960s, like welfare and other government programs, which he contended provided incentives for staying single

“When the economy recovers, you’ll still see all these problems persisting for reasons that have nothing to do with money and everything to do with culture,” he said.

There are no easy answers, in part because the problem is so complex, said Douglas J. Besharov, a fellow at the Atlantic Council. Blaming the problem on the richest of the rich ignores an equally important driver, he said: two-earner household wealth, which has lifted the upper middle class ever further from less educated Americans, who tend to be single parents.

The problem is a puzzle, he said. “No one has the slightest idea what will work. The cupboard is bare.”

Mentioned Publications

The widening academic achievement gap between the rich and the poor: new evidence and possible explanations.

Sean F. Reardon

In this chapter I examine whether and how the relationship between family socioeconomic characteristics and academic achievement has changed during the last fifty years. In particular, I investigate the extent to which the rising income inequality of the last four decades has been paralleled by a similar increase in the income achievement gradient. As the income gap between high- and low-income families has widened, has the achievement gap between children in high- and low-income families also widened?

The answer, in brief, is yes. The achievement gap between children from high- and low-income families is roughly 30 to 40 percent larger among children born in 2001 than among those born twenty-five years earlier. In fact, it appears that the income achievement gap has been growing for at least fifty years, though the data are less certain for cohorts of children born before 1970. In this chapter, I describe and discuss these trends in some detail. In addition to the key finding that the income achievement gap appears to have widened substantially, there are a number of other important findings.

First, the income achievement gap (defined here as the average achievement difference between a child from a family at the 90th percentile of the family income distribution and a child from a family at the 10th percentile) is now nearly twice as large as the black-white achievement gap. Fifty years ago, in contrast, the black-white gap was one and a half to two times as large as the income gap. Second, as Greg Duncan and Katherine Magnuson note in chapter 3 of this volume, the income achievement gap is large when children enter kindergarten and does not appear to grow (or narrow) appreciably as children progress through school. Third, although rising income inequality may play a role in the growing income achievement gap, it does not appear to be the dominant factor. The gap appears to have grown at least partly because of an increase in the association between family income and children's academic achievement for families above the median income level: a given difference in family incomes now corresponds to a 30 to 60 percent larger difference in achievement than it did for children born in the 1970s. Moreover, evidence from other studies suggests that this may be in part a result of increasing parental investment in children's cognitive development. Finally, the growing income achievement gap does not appear to be a result of a growing achievement gap between children with highly and less-educated parents. Indeed, the relationship between parental education and children's achievement has remained relatively stable during the last fifty years, whereas the relationship between income and achievement has grown sharply. Family income is now nearly as strong as parental education in predicting children's achievement.

This chapter is now published in the book Whither Opportunity: https://www.russellsage.org/publications/whither-opportunity

You are here

• Sean Reardon

• Chronicle Conversations
• Article archives
• Issue archives
• Join our mailing list

Reducing Poverty Through Education - and How

About the author, idrissa b. mshoro.

There is no strict consensus on a standard definition of poverty that applies to all countries. Some define poverty through the inequality of income distribution, and some through the miserable human conditions associated with it. Irrespective of such differences, poverty is widespread and acute by all standards in sub-Saharan Africa, where gross domestic product (GDP) is below $1,500 per capita purchasing power parity, where more than 40 per cent of their people live on less than $1 a day, and poor health and schooling hold back productivity. According to the 2009 Human Development Report, sub-Saharan Africa's Human Development Index, which measures development by combining indicators of life expectancy, educational attainment, and income lies in the range of 0.45-0.55, compared to 0.7 and above in other regions of the world. Poverty in sub-Saharan Africa will continue to rise unless the benefits of economic development reach the people. Some sub-Saharan countries have therefore formulated development visions and strategies, identifying respective sources of growth.

Tanzania case study

The Tanzania Development Vision 2025, for example, aims at transforming a low productivity agricultural economy into a semi-industrialized one through medium-term frameworks, the latest being the National Strategy for Growth and Reduction of Poverty (NSGRP). A review of NSGRP implementation, documented in Tanzania's Poverty and Human Development Report 2009, attributed the falling GDP -- from 7.8 per cent in 2004 to 6.7 per cent in 2006 -- to the prolonged drought during 2005/06. A further fall to 5 per cent was projected by 2009 due to the global financial crisis. While the proportion of households living below the poverty line reduced slightly from 35.7 per cent in 2000 to 33.6 per cent in 2007, the actual number of poor Tanzanians is increasing because the population is growing at a faster rate. The 2009 HDR showed a similar trend whereby the Human Development Index in Tanzania shot up from 0.436 to 0.53 between 1990 and 2007, and in the same year the GDP reached $1,208 per capita purchasing power parity. Again, the improvements, though commendable, are still modest when compared with the goal of NSGRP and Millennium Development Goal 1 to reduce by 50 per cent the number of people whose income is less than $1 a day by 2010 and 2015.

More deliberate efforts are therefore required to redress the situation, with more emphasis placed particularly on education, as most poverty-reduction interventions depend on the availability of human capital for spearheading them. The envisaged economic growth depends on the quantity and quality of inputs, including land, natural resources, labour, and technology. Quality of inputs to a great extent relies on embodied knowledge and skills, which are the basis for innovation, technology development and transfer, and increased productivity and competitiveness.

A quick assessment in June 2010 of education statistics in Tanzania indicated that primary school enrolment increased by 5.8 per cent, from 7,959,884 pupils in 2006 to 8,419,305 in 2010. The Gross Enrolment Ratio (GER) was 106.4 per cent. The transition rate from primary to secondary schools, however, decreased by 6.6 per cent from 49.3 per cent in 2005 to 43.9 per cent in 2009. On an annual average, out of 789,739 pupils who completed primary education, only 418,864 continued on to secondary education, notwithstanding the expansion of secondary school enrolment, from 675,672 students in 2006 to 1,638,699 in 2010, a GER increase from 14.8 to 34.0 percent. Moreover, the observed expansion in secondary school education mainly took place from grades one through four, where the number increased from 630,245 in 2006 to 1,566,685 students in 2010. As such, out of 141,527 students who on an annual average completed ordinary secondary education, only 36,014 proceeded to advanced secondary education. Some improvements have also been recorded at the tertiary level. While enrolment in universities was 37,667 students in 2004/05, there were 118,951 in 2009/10.

Adding to this number the students in non-university tertiary institutions totalled 50,173 in 2009/10 and the overall tertiary enrolment reached 169,124 students, providing a GER of 5.3 percent, which is very low.

The observed transition rates imply that, on average, 370,875 primary school children terminate their education journey every year at 13 to 14 years of age in Tanzania. The
17- to 19-year-old secondary school graduates, unable to obtain opportunities for further education, worsen the situation and the overall negative impact on economic growth is very apparent, unless there are other opportunities to develop and empower the secondary school graduates. Vocational education and training could be one such opportunity, but the total current enrolment in vocational education in Tanzania is about 117,000 trainees, which is still far from actual needs. A long-term strategy is therefore critical to expand the capacity for vocational education and training so as to increase the employability of the rising numbers of out-of-school youths. This fact was also apparent in the 2006 Tanzania Integrated Labour Force Survey, which indicated that youth between 15 and 24 years were more likely to be unemployed compared to other age groups because they were entering the labour market for the first time without any skills or work experience. The NSGRP target was to reduce unemployment from 12.9 per cent in 2000/01 to 6.9 per cent by 2010; hence the unemployment rate of 11 per cent in 2006 was disheartening.

One can easily notice that while enrolment in basic education is promising, the situation at other levels remains bleak in meeting poverty reduction targets. Moreover, apart from the noticeably low university enrolment in Tanzania, only 29 per cent of students are taking science and technology courses, probably due to the small catchment pool at lower levels. While this is so, sustainable and broad-based growth requires strengthening of the link between agriculture and industry. Agriculture needs to be modernized for increased productivity and profitability; small and medium enterprises, promoted, with particular emphasis on agro-processing, technology innovation, and upgrading the use of technologies for value addition; and all, with no or minimum negative impact on the environment. Increased investments in human and physical capital are also highly advocated, focusing on efficient and cost-effective provision of infrastructure for energy, information and communication technologies, and transport with special attention to opening up rural and other areas with economic potential. All these point to the promotion of education in science and technology. Special incentives for attracting investments towards accelerating growth are also emphasized. Experience from elsewhere indicates that foreign direct investment contributes effectively to economic growth when the country has a highly-educated workforce. Domestic firms also need to be supported and encouraged to pay attention to product development and innovation for ensuring quality and appropriate marketing strategies that make them competitive and capable of responding to global market conditions.

It is therefore very apparent from the Tanzania example that most of the required interventions for growth and the reduction of poverty require a critical mass of high-quality educated people at different levels to effectively respond to the sustainable development challenges of nations.  

The UN Chronicle  is not an official record. It is privileged to host senior United Nations officials as well as distinguished contributors from outside the United Nations system whose views are not necessarily those of the United Nations. Similarly, the boundaries and names shown, and the designations used, in maps or articles do not necessarily imply endorsement or acceptance by the United Nations.

What if We Could Put an End to Loss of Precious Lives on the Roads?

Road safety is neither confined to public health nor is it restricted to urban planning. It is a core 2030 Agenda matter. Reaching the objective of preventing at least 50 per cent of road traffic deaths and injuries by 2030 would be a significant contribution to every SDG and SDG transition.

Promoting Evidence-Based Prevention Strategies to Mitigate the Harms of Drug Use: The Role of the United Nations Office on Drugs and Crime

The engagement of the United Nations Office on Drugs and Crime with Member States is particularly focused on interventions addressing early adolescence through schools and families by piloting evidence-based, manualized programmes worldwide. 

A Chronicle Conversation with Pradeep Kurukulasuriya (Part 2)

In April 2024, Pradeep Kurukulasuriya was appointed Executive Secretary of the United Nations Capital Development Fund (UNCDF). The  UN Chronicle  took the opportunity to ask Mr. Kurukulasuriya about the Fund and its unique role in implementing the 2030 Agenda for Sustainable Development. This is Part 2 of our two-part interview.

Documents and publications

• Yearbook of the United Nations 
• Basic Facts About the United Nations
• Journal of the United Nations
• Meetings Coverage and Press Releases
• United Nations Official Document System (ODS)
• Africa Renewal

Libraries and Archives

• Dag Hammarskjöld Library
• UN Audiovisual Library
• UN Archives and Records Management 
• Audiovisual Library of International Law
• UN iLibrary 

News and media

• UN News Centre 
• UN Chronicle on Twitter
• UN Chronicle on Facebook

The UN at Work

• 17 Goals to Transform Our World
• Official observances
• United Nations Academic Impact (UNAI)
• Protecting Human Rights
• Maintaining International Peace and Security
• The Office of the Secretary-General’s Envoy on Youth
• United Nations Careers

• Skip to main content
• Keyboard shortcuts for audio player

Congress poured billions of dollars into schools. Did it help students learn?

Cory Turner

2 reports set out to answer whether K-12 students have recovered from the pandemic

Two new studies offer a first look at how much more students learned thanks to federal pandemic aid money. Blend Images - JGI/Jamie Grill/Tetra images RF/Getty Images hide caption

America’s schools received an unprecedented $190 billion in federal emergency funding during the pandemic. Since then, one big question has loomed over them: Did that historic infusion of federal relief help students make up for the learning they missed?

Two new research studies, conducted separately but both released on Wednesday, offer the first answer to that question: Yes, the money made a meaningful difference. But both studies come with context and caveats that, along with that headline finding, require some unpacking.

How much of a difference did the money make?

$190 billion is an enormous amount of money by any measure. But districts were only required to spend a fraction of the relief on academic recovery, by paying for proven interventions like summer learning and high-quality tutoring. So how much additional student learning did the federal aid actually buy?

Study #1 , a collaboration including Tom Kane at Harvard’s Center for Education Policy Research and Sean Reardon at Stanford’s Educational Opportunity Project, estimates that every $1,000 in federal relief spent per student bought the kind of math test score gains that come with 3% of a school year, or about six school days of learning. That’s during the 2022-23 academic year.

Pandemic aid for schools is ending soon. Many after-school programs may go with it

Improvements in reading scores were smaller: roughly three school days of progress per $1,000 in federal relief spending per student.

The federal relief “was worth the investment,” Reardon tells NPR. “It led to significant improvements in children's academic performance… It wasn't enough money, or enough recovery, to get students all the way back to where they were in 2019, but it did make a significant difference.”

Study #2 , co-authored by researcher Dan Goldhaber at the University of Washington and American Institutes for Research, offers a similar estimate of math gains. The increase in reading scores, according to Goldhaber, appeared comparable to those math gains, though he says they’re less precise and a little less certain.

“It did have an impact,” Goldhaber tells NPR, an impact that’s “in line with estimates from prior research about how much money moves the needle of student achievement.”

Who benefited the most?

The federal recovery dollars came in three waves, known as ESSER ( Elementary and Secondary School Emergency Relief Fund ) I, II and III. The first two waves were relatively small, roughly $68 billion, compared to the $122 billion of ESSER III.

The windfall was distributed to schools based largely on need – specifically, based on the proportion of students living in or near poverty. The assumption being: Districts with higher rates of student poverty would need more help recovering. COVID hit high-poverty communities harder, with higher rates of infection, death, unemployment and remote schooling than in many affluent communities.

Homelessness in the U.S. hit a record high last year as pandemic aid ran out

“These and other factors likely caused greater learning loss during the pandemic and dampened academic recovery,” Goldhaber writes in Study #2, pointing out that, “the Detroit, MI public school district received about $25,800 per pupil across all waves of ESSER… [while] Grosse Pointe, MI (a nearby suburb) only received about $860 per pupil.”

Here’s where the story of these federal dollars gets complicated, because the learning they appear to have bought wasn’t experienced evenly, according to Goldhaber.

In Study #2, he and co-author Grace Falken, found larger academic benefits from federal spending in districts serving low shares of Black and Hispanic students. Though he tells NPR, these patterns “do not necessarily imply that ESSER's impacts vary because of student demographics. Rather, the results could reflect other district characteristics that happen to correlate with the student populations the districts serve.”

Reardon and Kane did not find statistically significant evidence of this kind of variation.

Goldhaber and Falken also found that towns saw more math gains than cities, while rural areas led the way in reading growth. Interestingly, suburban districts generally experienced “smaller, insignificant impacts” from the federal spending in both subjects.

But did the money help enough?

If your standard for “enough” is a full recovery for all students from the learning they missed during the pandemic, then no, the money did not remedy the full problem.

But the researchers behind both studies say that’s an unrealistic and unreasonable yardstick. After all, Congress only required that districts spend at least 20% of ESSER III funds on learning recovery. The rest of the relief came with relatively few strings attached.

Kindergartners are missing a lot of school. This district has a fix

Instead, the researchers say, the money’s effectiveness should be judged by a more realistic standard, based on what previous research has shown money can and cannot buy.

Harvard’s Tom Kane, of Study #1, points out that their results do line up with pre-pandemic research on the impact of school spending, and suggest a clear, long-term return on investment.

“These academic gains will translate into improvements in earnings and other outcomes that will last a lifetime,” Kane tells NPR.

For example, the academic gains associated with every $1,000 in per student spending would be worth $1,238 in future earnings, Kane estimates. Increased academic achievement also comes with valuable social returns, he says, including lower rates of arrest and teen motherhood.

What’s more, Reardon tells NPR, because these federal dollars disproportionately went to lower-income districts, “not only do we find that the federal investment raised test scores, but we also find that it reduced educational inequality.”

But the work’s not over.

In Study #2, Goldhaber and Falken write, “to recover from these remaining losses, our estimates suggest schools would need between $9,000 and $13,000 in additional funds per pupil, assuming the return on those funds is similar to what we estimated for ESSER III.”

They also warn that middle-income districts could continue to struggle – because they experienced academic losses but got less federal aid.

In a presidential election year, it’s unlikely Congress will agree to send schools more money. And Goldhaber worries, as ESSER funds begin to expire this year, districts will have to cut staff.

“Some districts, particularly high poverty, high minority districts, are going to lose so much money that I think teacher layoffs are inevitable,” Goldhaber tells NPR. “So I'm worried that the funding cliff – there's a downside that we're not thinking hard enough about.”

The good news, says Kane, is that ESSER was a massive, “brute force” effort, and a far smaller, state-driven effort could still make a big difference, so long as it’s hyper-focused on academic interventions.

Kane says, “It falls to states to complete the recovery.”

How Federal Pandemic Aid Impacted Schools

• Posted June 26, 2024
• By Elizabeth M. Ross
• Disruption and Crises
• Education Finances
• Education Policy
• Education Reform
• Student Achievement and Outcomes

K–12 schools received nearly $190 billion in federal relief during the COVID-19 pandemic, 90% of which went directly to local districts. Financially disadvantaged districts received the most aid money, but how effective was the money at helping students make up the learning they missed during the pandemic?

Answers can be found in new research which measured the impact of the spending by looking at the average test scores in reading and math from the spring of 2022–2023, for students in grades 3–8. The researchers were not able to assess which intervention strategies were the most effective because school districts were not required to report how they spent the funds they received.

Professor Thomas Kane , economist and co-author of the new report from the Center for Education Policy Research at Harvard University and The Educational Opportunity Project at Stanford University, explains the role that federal relief money played in the academic recovery story in 29 states.

Could you summarize what you found out about the impact of the federal relief money on student achievement during the 2022 to 2023 school year?

We found that $1,000 of federal aid per student that a district spent during the 2022–2023 school year was associated with a 0.03 grade equivalent rise in math achievement (or approximately 6 days of learning) and in reading, the effects were somewhat smaller, a 0.018 grade equivalent (or approximately 3 days of learning). So, the effects were not huge. I think readers might look at that and say, oh gosh, that's a small effect. But what people don't realize is just how strongly related to longer-term outcomes test scores are. So, although the impacts per dollar spent were not large, given the relationship between K–12 test scores and earnings later in life, our estimates imply they were large enough to justify the investment. 

In the conclusion of your report, you say that the average recovery was actually larger than what you expected based on your estimate of the effect of the spending. Why was that? 

We were surprised when we first got the 2022–2023 data and saw the total magnitude of the gains that year. They were 170% as large as the average annual improvement during the last period of rapid growth in achievement, between 1990 and 2013, in math and double the improvement in reading during that time period.

In this report, we investigated the role that the federal aid played in that growth. Our primary challenge was sorting out how much of the growth was due to spending, versus how much of the growth was related to community poverty — since poorer districts received more aid on average. We took several different approaches to doing that — for instance, using state differences in the Title I formula on which the funding was based and finding high-poverty districts which received large grants (because of the state they were in or because of anomalies in the aid formula) and similarly high-poverty districts with much smaller grants but similar prior trends in achievement. We tried multiple approaches and found similar answers each way we looked at it.

We're still surprised, partially because of the news over the past few years of districts spending the federal relief on athletics fields and across-the-board pay raises and the implementation challenges districts faced when trying to implement tutoring or recruiting students to summer school. But the dollars seem to have had an impact.

"Imagine if, at the beginning of the pandemic, the federal government did not even try to coordinate efforts to develop a vaccine. Instead, suppose they took all that money and sent it to local public health departments, saying, 'You figure it out.' Some would have succeeded, but many would have failed. That’s exactly what happened in the K–12 response."  Professor Thomas Kane

In your report you suggested that parental help at home, efforts on the part of teachers and students, and possibly increases in spending at the local level may have played a role in the recovery effort. And it's interesting because I remember the last time I talked with you , you mentioned your concerns about the lack of coordination with the spending of the federal relief money. Is that still a concern? 

Yes, in some ways the federal aid was like the first stage of a rocket — it got us started but was broadly focused and ultimately insufficient to get us all the way there. Part of that was due to a lack of coordination. Each district was developing and implementing plans largely on their own. It could have been much more effectively spent. For instance, research suggests that the cost effectiveness ratio for a high-dosage tutoring program was roughly 10 times as large as the cost effectiveness we found for each $1000 in aid spent.  

In the report, we also recommend efforts states should be doing now to continue the recovery, because it's pretty clear that there won't be another federal package, given what's happening in Washington. It’s alarming, but it’s just not on the radar screen of most governors — including here in Massachusetts, where the highest-poverty districts have actually lost additional ground since the pandemic. States have spent the last few years watching districts spend down their federal pandemic relief dollars, not recognizing that the recovery will not be complete when the federal dollars run out. Simply going back to business as usual will leave a lot of our neediest communities further behind than they were before the pandemic. So, we're hoping that these results become a call to action at the state and local level. It’s in governors and state legislators’ hands now. If they don’t step up, poor children will end up bearing the most inequitable and longest lasting burden from the pandemic. 

The aid did, by our estimates, seem to have a disproportionate effect on high-poverty districts, mostly because they got a lot more money. But that wasn't enough to completely offset the losses. The highest-poverty districts remain behind as well as the middle-income districts. The wealthiest districts we anticipate will be back to 2019 levels soon, not because they received much federal aid — they did not — but because they did not fall very far behind in the first place.

Are there lessons to be learned overall from the pandemic recovery effort? 

I do think it would have been beneficial to give federal regulators and state governments more opportunities to coordinate local efforts — like to plan statewide tutoring programs or to plan statewide summer learning programs. Most of the bigger districts would have had the staff to plan their own efforts, but the medium and smaller districts, they didn't necessarily have the bandwidth to be thinking about planning for major summer learning initiatives and tutoring programs. I think granting states, and the federal government, more say in approving local recovery plans, in ensuring that what districts were planning were sufficient to help students catch up and giving states more money to coordinate efforts would have helped. 

Imagine if, at the beginning of the pandemic, the federal government did not even try to coordinate efforts to develop a vaccine. Instead, suppose they took all that money and sent it to local public health departments, saying, “You figure it out.” Some would have succeeded, but many would have failed. That’s exactly what happened in the K–12 response. 90% of the federal aid went directly to local school districts. Some figured it out, but many did not.

States and districts should have plans on the shelf for what happens in the next pandemic. I'm sure there are individual schools that will say that they know exactly what they would do next time. But there has not been that sort of learning at the state level — since most states just took a back seat. I have not heard much planning at the state or federal level about what they would do differently next time — and how they might plan for a major tutoring initiative or assembling materials for summer learning, etc. We're not going to have better coordination next time unless somebody starts planning now. 

The latest research, perspectives, and highlights from the Harvard Graduate School of Education

Related Articles

Where Have All the Students Gone?

New Research Provides the First Clear Picture of Learning Loss at Local Level

Despite progress, achievement gaps persist during recovery from pandemic.

New research finds achievement gaps in math and reading, exacerbated by the COVID-19 pandemic, remain and have grown in some states, calls for action before federal relief funds run out

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 03 July 2024

Geographical migration and fitness dynamics of Streptococcus pneumoniae

• Sophie Belman   ORCID: orcid.org/0000-0002-9778-7174 1 , 2 , 3 ,
• Noémie Lefrancq   ORCID: orcid.org/0000-0001-5991-6169 2 ,
• Susan Nzenze 4 ,
• Sarah Downs   ORCID: orcid.org/0000-0001-5871-5373 5 ,
• Mignon du Plessis   ORCID: orcid.org/0000-0001-9186-0679 6 , 7 ,
• Stephanie W. Lo   ORCID: orcid.org/0000-0002-2182-0222 1 , 8 ,
• The Global Pneumococcal Sequencing Consortium ,
• Lesley McGee 9 ,
• Shabir A. Madhi   ORCID: orcid.org/0000-0002-7629-0636 5 , 10 ,
• Anne von Gottberg 6 , 7 , 11 ,
• Stephen D. Bentley   ORCID: orcid.org/0000-0001-8094-3751 1   na1 &
• Henrik Salje   ORCID: orcid.org/0000-0003-3626-4254 2   na1  

Nature ( 2024 ) Cite this article

Metrics details

• Bacterial genetics
• Population dynamics

Streptococcus pneumoniae is a leading cause of pneumonia and meningitis worldwide. Many different serotypes co-circulate endemically in any one location 1 , 2 . The extent and mechanisms of spread and vaccine-driven changes in fitness and antimicrobial resistance remain largely unquantified. Here using geolocated genome sequences from South Africa ( n  = 6,910, collected from 2000 to 2014), we developed models to reconstruct spread, pairing detailed human mobility data and genomic data. Separately, we estimated the population-level changes in fitness of strains that are included (vaccine type (VT)) and not included (non-vaccine type (NVT)) in pneumococcal conjugate vaccines, first implemented in South Africa in 2009. Differences in strain fitness between those that are and are not resistant to penicillin were also evaluated. We found that pneumococci only become homogenously mixed across South Africa after 50 years of transmission, with the slow spread driven by the focal nature of human mobility. Furthermore, in the years following vaccine implementation, the relative fitness of NVT compared with VT strains increased (relative risk of 1.68; 95% confidence interval of 1.59–1.77), with an increasing proportion of these NVT strains becoming resistant to penicillin. Our findings point to highly entrenched, slow transmission and indicate that initial vaccine-linked decreases in antimicrobial resistance may be transient.

The greatest public health burden from infectious diseases remains stubbornly endemic pathogens. Once established, pathogens such as Mycobacterium tuberculosis , HIV and, now, SARS-CoV-2 are difficult to control, even when vaccines are available 1 . Their persistence in the population can be partially explained by the co-circulation of multiple strains of the same pathogen. Endemic pathogens are complicated to study as we rarely understand the mechanisms that drive spread, including the role of human behaviour and why some lineages increase in prevalence over time whereas others disappear. Underlying genetic diversity is particularly extreme in the case of the bacterium S.   pneumoniae (the pneumococcus), which is the leading cause of morbidity and mortality worldwide because of lower respiratory infections 2 , 3 , 4 . The pneumococcus comprises >100 known antigenically distinct serotypes and >900 classified lineages (also known as global pneumococcal sequence clusters (GPSCs)) 5 , 6 , 7 . Moreover, it is not uncommon for more than 30 antigenically distinct serotypes to co-circulate within a country or region or for a human host to concurrently carry multiple serotypes 8 . We refer to lineage synonymously with GPSC, whereas a strain references any particular circulating phenotype (including specific serotypes and antimicrobial resistance (AMR)). Here we develop mathematical models using thousands of geolocated genome sequences from South Africa collected over a 15-year period to clarify several key uncertainties in pneumococcal migration. We model the rate and breadth of mobility geographically and how fitness changes linked to vaccine implementation and AMR may affect its spread.

The pneumococcus resides in the human upper respiratory tract. Carriage is a prerequisite for disease, and rates of carriage in children under 5 years old range from 20 to 90% 9 . Occasionally, asymptomatic carriage goes on to cause local infections such as otitis media or, more severely, invasive pneumonia and meningitis. More than 500,000 deaths per year linked to pneumococcus are estimated to occur globally 3 , 10 . Penicillin was first used to treat pneumococcal disease in the 1930s, and it successfully reduced pneumococcal disease until the late 1960s when penicillin non-susceptibility was first noted. Multidrug-resistant strains were described soon after penicillin resistance 11 , 12 . By 2019, 19% of deaths associated with AMR had pneumococcal aetiology 13 . In this context, vaccines are pivotal to disease control. A pneumococcal polysaccharide vaccine that included 23 serotypes was licensed in the USA in 1983. However, the absence of mucosal immunity has seen it replaced by pneumococcal conjugate vaccines (PCVs) except for in older adults and immunocompromised individuals 14 . PCVs (conjugated with toxin to stimulate mucosal immunity) target a small subset of the polysaccharide capsular serotypes, with the most common formulations including PCV7 and PCV13 (Pfizer) 15 and PCV10 (GlaxoSmithKline) 16 . These target 7, 13 and 10 serotypes, respectively (with all the serotypes included within PCV7 and PCV10 also included within PCV13). In 2021, additional PCV formulations, PCV15 (Merck) and PCV20 (Pfizer), were licensed for use in the USA and Europe 17 . Pneumococcal vaccination is dynamic, and new vaccine compositions are frequently tested. Vaccine serotypes are often selected because of their high prevalence and AMR among disease isolates from infants and children. PCVs are now included in 76% of national immunization schedules, with different formulations in different countries. For example, in South Africa, PCV7 was implemented in 2009 and replaced by PCV13 in 2011, excluding PCV10 (ref. 18 ). Despite their success at reducing disease, their use has been linked to serotype replacement by NVTs in both invasive pneumococcal disease (IPD) and carriage 8 , 19 , 20 , 21 . In South Africa, this has been characterized by increases in NVT serotypes 8 and 15B among IPD, and increases in NVT serotypes 16F, 24, 35B and 11A among carriage isolates (8, 15B and 11A are now included in PCV20) 20 , 22 , 23 . Although there has been success in predicting the fitness of individual isolates based on the overall gene distribution in a population, quantitative measures of fitness linked to the serotype of individual isolates are lacking 24 , 25 . This includes quantifying the time it takes after implementation for vaccines to affect the serotype composition in the country. Moreover, quantifying the serotype growth before and after vaccine implementation at different time points is needed. These are crucial knowledge gaps, as serotype distributions ultimately drive vaccine development and deployment strategies. In addition, vaccine implementation has resulted in reductions in AMR among both IPD and carriage isolates 20 , 22 , 26 . However, it remains unclear whether these reductions will persist over time at the population level or whether AMR may rebound.

Mathematical models applied to geolocated pathogen genome sequence data are useful to disentangle the changing prevalence of different lineages. However, most phylogeographical models focus on the rate of pathogen flow between locations, which represents the overall effect of multiple transmission chains. Such models therefore consider a different ecological scale to the specific behaviours of infected people and the surrounding population at each transmission generation 27 . The relationship between behaviours of individuals at each transmission step and the overall patterns of pathogen flow between locations are complex and nonlinear. Most existing phylogeographical approaches also struggle to account for changing levels of surveillance in both space and time. Here we develop mechanistic models that use the generation time distribution to estimate the number of transmission events that separate the most recent common ancestors (MRCAs) from each pair of tips in a time-resolved phylogeny. Taken together with measures of human mobility probabilities and human population distribution, we can infer mechanisms of pneumococcal migration at each transmission generation. We implemented this model with a focus on South Africa, where approximately 65% of children ≤5 years of age (35% across all age groups) carry the pneumococcus 28 , 29 . We incorporate both uncertainty in the phylogenetic reconstructions and sampling uncertainty through a bootstrapping approach 30 . We explore the robustness of our approach to highly biased observation processes using simulated data with known parameter values. Finally, we quantify the changing fitness of strains in response to vaccine introduction, including those containing AMR.

Quantifying spatial structure

In partnership with the South African National Institute for Communicable Disease and the Wits Vaccines and Infectious Diseases Analytics Research Unit, we sequenced the whole genomes of isolates from each of South Africa’s nine provinces between 2000 and 2014 ( n  = 6,910, 5,060 from individuals with invasive pneumococcal disease and 1,850 from carriage studies) (Fig. 1a–c ). Despite the large number of sequences, this dataset only represents a very small proportion ( ≪ 0.1%) of the circulating pneumococcus over this time period. We identified 184 GPSCs with 69 different serotypes (31.9% NVT) (Fig. 1a,e and Supplementary Table 1 ). This diversity persisted across provinces, and the distribution of serotypes within GPSCs did not follow a distinct geographical structure (Fig. 1a ). In silico predicted AMR was common (penicillin, 48.2%; erythromycin, 17.3%, clindamycin, 11.2%; and co-trimoxazole: 68.4%), with similar distributions for isolates from both carriage and disease (Fig. 1f , Extended Data Fig. 10 , Supplementary Fig. 13 and Supplementary Table 3 ).

a , Phylogenetic tree of 6,910 South African isolates included in this study. Dominant GPSCs ( n  > 50) are in purple. GPSC1 (top) and GPSC5 (bottom) are highlighted. The columns describe the serotypes and provincial region for each isolate. The branch length legends refer to single nucleotide polymorphisms (SNPs) per site and trees are midpoint rooted. b , Map of the nine provinces of South Africa coloured by province. Scale bar is included in kilometres (km).  c , Count of isolates ( n  = 6,910) per collection year from 2000 to 2014 used in the lineage-level analysis (black) and the 9 dominant GPSCs used in the divergence time analysis (maroon). d , The mean geographical distance for sequence pairs as a function of cumulative evolutionary distance across all GPSCs with 95% CI (blue). The model fit is shown in red. The implied true pattern of spread is shown in purple, after accounting for a biased observation process. e , The proportion of NVT serotypes across the study period. f , The proportion of in silico predicted AMR isolates for four drugs across the study period. The vertical lines denote the introduction of PCV7 in 2009 and PCV13 in 2011. An interactive phylogeny and metadata are available at Microreact ( https://microreact.org/project/7wqgd2gbBBEeBLLPKonbaT-belman2024southafricapneumococcus ).

Taking the 9 most dominant GPSCs in turn (each comprising more than 50 sequences, 2,575 sequences in total), we built recombination-free, time-resolved phylogenetic trees to determine the divergence times between sequence pairs (Extended Data Fig. 1a–i and Supplementary Table 4 ). We included 1,157 genomes from 14 other countries in Africa and 2,944 from 31 countries outside Africa in our phylogenies. We compared the geographical distance spread per divergence time between pairs in South Africa and found a clear geographical structure. The geographical distance between pairs increased from a mean distance of 142 km (95% confidence interval (CI) = 54–207 km) for those separated by less than 2 years of evolutionary time to 297 km (95% CI = 274–323 km) for those separated by 10–20 years (Fig. 1d ). We obtained consistent results when using only sequences that came from individuals with disease (Supplementary Fig. 1A and Supplementary Table 5 ) and across the different GPSCs (Supplementary Fig. 2 ). This result is consistent with largely common patterns of spread irrespective of which particular GPSC an individual is infected with. Despite high heterogeneity in GPSC composition within any province, overall, pairs of isolates that are from the same province had 1.27 (95% CI = 1.23–1.31) times the relative risk (RR) of being the same GPSC compared with pairs of isolates from distant provinces (>1,000 km). This RR fell to 1.09 (95% CI = 1.03–1.12) for pairs separated by 500–1,000 km (Fig. 2a and Supplementary Table 5 ). We obtained consistent results when we limited our analysis to only disease isolates and when we subsampled to have even numbers of sequences by province to mitigate sampling bias (Extended Data Fig. 2a,b and Supplementary Table 5 ). As there can be hundreds of years of diversity within a single GPSC, we refined the analysis by using different evolutionary windows of separation between pairs of isolates as determined from the phylogenetic trees. Pairs from the same province had 3.87 (95% CI = 3.01–5.1) times the RR of having a recent common ancestor (within 5 years) than distal pairs (>1,000 km apart) (Fig. 2b and Supplementary Table 5 ). However, as the evolutionary time between isolates increases the relative probability of being from the same province decreases, it is only after around 50 years that the pneumococcus seemed to be well mixed throughout the country (Fig. 2c–f and Supplementary Table 6 ). Furthermore, comparisons of the spatial location of closely related pairs showed that pneumococcus flow was dominated by within-country movement compared with either movement between South Africa and other African countries or non-African countries (Fig. 2c–e , Supplementary Fig. 3A and Supplementary Table 5 ). Recognizing that our samples span age groups, we stratified the analysis by the age difference between pairs. Genome pairs from individuals who were greater than 5 years apart in age took slightly longer to become mixed across South Africa (Supplementary Fig. 3B and Supplementary Table 7 ). These findings are consistent with a highly entrenched pathogen that moves slowly within a country and with slow cross-border transmission.

a , RR of being the same GPSC within a province (blue), between different provinces over increasing distance (red) and compared with geographically distant pairs (>1,000 km) (reference). (South Africa; n  = 6,910). b – e , RR of having a time to most recent common ancestor (tMRCA) 0–5 years ( b ), 5–10 years ( c ), 10–20 years ( d ) and 20–200 years ( e ) ago within South African provinces (blue), across larger distances within South Africa (red), from South Africa to other countries in Africa ( n  = 1,157) (green), and from South Africa to countries outside of Africa ( n  = 2,944) (purple). All plots use a reference of pairs that are from distant provinces in South Africa (open triangle). f , RR of similarity over rolling 20-year windows of divergence times for pairs isolated within the same South African province compared with pairs from distant provinces in South Africa (>1,000 km apart). For a – f , plots are centred at the median and error bars represent 2.5 and 97.5 percentiles across posterior phylogenies.

Inferring migration using human mobility

To understand whether human mobility can explain the slow spread of the pneumococcus, we built a mechanistic model of geographical spread fit to the nine dominant GPSCs and the observed province in which our genome sequences were isolated. We used the generation time distribution, time from one person being infected to infecting the next person (estimated mean of 35 days, standard deviation of 35 days (gamma distribution)) to translate branch lengths to the number of generations between pairs of sequences 31 , 32 (Supplementary Fig. 4 ). Each transmission generation is an opportunity for pneumococcal mobility. We used directional human mobility probabilities between each of the 234 South African municipalities from Meta Data for Good 30 to infer the probable location of a single transmission event, allowing for mobility of both the infected individual and the surrounding susceptible population. As Meta data captures mobility over a single day, we adjusted the duration in any cell to consider total mobility over the infectious period. Furthermore, as Meta users may act differently to those involved in pneumococcal transmission, we incorporated a parameter that allows individuals to have a different probability of staying within their home municipality than Meta users (Fig. 3b ). We then calculated the probability of pneumococcal movement between each pair of locations for each transmission generation. This approach integrates over all possible pathways linking two locations. We incorporated the probability of an isolate being sequenced at each geographical location and within each collection year to account for the number of isolates being sequenced differing by year and location. We fit the model in a Bayesian framework using Markov chain Monte Carlo (MCMC). We compared the performance of our models using the Meta data to a gravity model in which the probability of mobility is a function of the distance to a location and its population size (Extended Data Fig. 4 ). We also separately modelled mobility that depended on distance only.

a , The estimated probability of the location (province) of each MRCA ( y  axis) compared with the population size ( x  axis) in that province. Points are centred at the median and error bars represent 95% credible intervals. b , The proportion of individual or pathogen mobility as a function of distance from the origin location. We compared the mean distance travelled when we consider the infector only (crossed square), when we consider the mobility of both the infector and the infectee (black filled circle) after a single transmission generation, as well as the overall mobility of the pathogen after ten generations (triangle). As a comparison, we present the expected pathogen spread after a single generation if transmission was completely spatially random (maroon). We also present the difference between Meta users (blue circle) and the movement of those involved in transmission. Points are centred at the median and error bars represent 95% credible intervals. c , The RR of being in each of the 234 municipalities of South Africa after 1 year (10 transmission generations) of sequential person-to-person transmission compared with being in a randomly selected municipality. Black dots denote municipalities with populations of >3 million people. d , The number of unique municipalities visited for 500 unique sequential simulations (grey) and the mean (black-dashed) across years of transmission following an introduction in a randomly selected municipality given the modelled migration probabilities at each transmission generation. e , The relative number of unique municipalities visited by NVT serotypes compared with VT serotypes after 1 and 2 years of transmission. Points are centred at the median and error bars represent 95% credible intervals.

Both the gravity model and the model that relied on Meta data were able to recover the observed spatial spread in pneumococcus (Fig. 1d , red line). However, model fit was better for the Meta data model (difference in deviance information criterion (DIC) of 3.4). Both these models outperformed the distance-only model (Supplementary Table 8 ). The Meta data model showed that the relative proportion of the population carrying the pneumococcus per province at any time is strongly correlated with the population size in that province ( R 2  = 0.97, P  ≤ 0.01; Fig. 3a ). This model enabled us to infer the true underlying spread of pneumococcus (that is, in which we accounted for the biased observation process and mobility in both the infected individual and the surrounding susceptible population) (Fig. 1d, dashed line). We estimated that among individuals involved in pneumococcal transmission, the daily probability of staying in their home municipality was 94.3% (95% CI = 93.8–95.0%) (Fig. 3b , black squares) compared with 99.8% (ranging from 86.6% in Mogale City, Gauteng, to 99.9% in Ba-Phalaborwa, Limpopo) for Meta users (Fig. 3b , blue points). When we incorporated the mobility of both infector and infectee and accounted for the mobility across the infectious period, we estimated that after a single transmission generation (35 days), 53.1% (95% CI = 46.6–58.2) of strains remained in their starting municipality, 22.8% (95% CI = 19.2–27.1%) were in a neighbouring municipality and a small minority were more than 500 km away (Fig. 3b , black points). As the number of transmission generations increased, the probability of reaching distal municipalities also increased (Fig. 3b ). The size of the community seemed to be key to determining where lineages travel. After 1 year of sequential transmission, the probability of being in a municipality with a population size of >3 million people was 26.7 (95% CI = 19.8–40.10) times that of being in a randomly selected municipality. This result is consistent with most pathogen movement passing through urban centres (Fig. 3c and Supplementary Table 9 ).

The municipality in which a strain emerges also seemed important. After 1 year of sequential transmission, a new strain that first occurred in a rural municipality (population density of <50 people per km 2 ) has travelled a median distance of 468.7 km (95% CI = 71.3–1,204.4 km), whereas in the same time window, a variant first occurring in an urban municipality (>500 people per km 2 ) has travelled only 285.6 km (95% CI = 36.3–967.0 km). Furthermore, the variant that emerged in a rural municipality would have travelled to 1.53 times as many municipalities as the urban variant (Extended Data Fig. 3a–c ). This result is corroborated by previous research that demonstrated high levels of in and out migration among individuals in rural settings owing to travel for work or education 33 , 34 . On average, 1 year after emerging, transmission chains visited 4 (95% CI = 1–8) municipalities, and after 10 years they visited 20 (95% CI = 13–27) municipalities (Fig. 3d ). Overall, these results show that the breadth of geographical spread is driven by a small number of long-range transmission events, with most transmissions remaining local. Incorporating our model into a branching epidemic, we found that after 10 years, transmission events are an average 465 km (95% CI = 456–472 km) from where they began (Supplementary Fig. 5A,B ).

To test the performance of our model, we simulated transmission within and between districts using the Meta matrix adjusted by a known parameter that determines the probability of staying within one’s home district. We then fit our model using a subset of infections to re-estimate this parameter. Even when only a small minority of infections were sequenced, we recovered the true probability of staying within a district. Our estimates were even robust to an extremely biased observation scenario whereby data from only two locations were available. Using this framework, we also explored the effect of misspecification of the generation time distribution. Using a 50% shorter generation time led to a small overestimate in the probability of staying in one’s home location each day (96.7% versus 95.3%), whereas using a 50% longer generation time had the opposite effect (Extended Data Fig. 5 ).

Vaccine-induced fitness changes

Implementation of PCV7 in 2009 and PCV13 in 2011 was associated with a substantial disruption in the patterns of circulating serotypes, results that are consistent with what has been previously observed 8 , 20 , 35 , 36 . However, the introduction of vaccines was also associated with a marked change in fitness. By 2014, serotypes included in PCV13 represented 33.2% of all isolates in our dataset, a reduction from 85.0% in the pre-vaccine era (Fig. 4a ). These patterns were consistent across the nine provinces in South Africa (Supplementary Fig. 6 ). To quantify changes in fitness linked to the vaccines, we fitted models to the annual distribution of serotypes across 184 GPSCs from the full dataset, allowing for differential fitness in serotypes included in PCV7 (serotypes 4, 6B, 9V, 14, 19F, 18C and 23F), PCV13 (which includes additional serotypes 1, 3, 5, 6A, 7F and 19A), and those not included in the vaccine (NVT). This method tracks the proportion of all serotypes at the population level over time and quantifies the relative advantage of each of them following the implementation of vaccines. This simple formulation was able to recover the observed distribution of serotype proportions in each group, by year, across provinces (Fig. 4a–c and Supplementary Fig. 6 ). We note that the number of NVT and VT isolates we used in our model do not represent the underlying incidence of NVT and VT, as only a small proportion of all infections were detected and sequenced. However, as we focused on the relative abundance of NVT and VT strains per year, our approach is robust to changes in the absolute numbers of isolates sequenced.

a – c , Data (points) and model fit (lines) for the proportion of serotypes from NVTs ( a ), PCV7 types ( b ) and additional PCV13 types not included in PCV7 ( c ) from the years 2000 to 2014 in this study. The long dashed line indicates the time of PCV7 implementation (2009) and the short, dashed line indicates the time of PCV13 implementation (2011). d , Relative fitness for the three groups of serotypes compared with the NVT fitness estimates before and after PCVs were introduced. e , Relative fitness estimates for all three groups of serotypes comparing the before and after PCV eras. For a – e , before PCV refers to before 2009 for NVT serotypes, before 2009 for PCV7 type serotypes and before 2011 for PCV13 type serotypes. f , Proportion of penicillin resistance overall (black line), within NVT strains (maroon points) and within VT strains (turquoise points) with model fits. The dashed line indicates the time of PCV implementation (2009). g , Relative fitness of penicillin resistance among NVTs (pink) and VTs (blue) in before (left) and after (right) PCVs. Data in d , e and g are on a log scale. For a – f , plots are centred at the median and include error bars representing 95% credible intervals around the posterior parameter distributions ( n  = 6,798).

Before the implementation of vaccines (2000–2008), NVTs had a relative fitness of 1.08 (95% CI = 1.06–1.09) compared with serotypes included in the vaccine (combining serotypes in PCV7 and PCV13 into VT). Following the implementation of PCV7 and PCV13, the fitness of the serotypes they target declined respectively compared with NVT serotypes (Fig. 4d and Supplementary Tables 10 and 11 ). When comparing serotype fitness before and after the introduction of PCVs, VTs had a relative fitness of 0.86 (95% CI = 0.78–0.92) and 0.76 (95% CI = 0.67–0.84) for PCV7 and PCV13 serotypes, respectively (Supplementary Table 11 ). Meanwhile, for the NVTs, vaccines were associated with a 1.25 (95% CI = 1.14–1.35) times increase in relative fitness from 2009 to 2014 compared with before the implementation of vaccines (Fig. 4e and Supplementary Tables 10 and 11 ). When we directly compared the fitness advantage of NVTs compared with VTs in the PCV era, NVTs had a relative fitness advantage of 1.68 (95% CI = 1.59–1.77), which is equivalent to a 1.05 (95% CI = 1.05–1.06) growth advantage (relative to VTs) at each transmission generation. In a sensitivity analysis, the results remained consistent for carriage and disease isolates, respectively, despite them having been sampled from different cohorts (Extended Data Fig. 6 and Supplementary Fig. 7 ). We also found consistent results across provinces (Supplementary Fig. 6 ). We additionally assessed whether there was a delay between vaccine implementation and resulting changes in strain fitness (Extended Data Fig. 7 ). The best fitting model assumed the change in fitness occurred in the same year as vaccine implementation.

Refining this model to look at the fitness of individual serotypes showed a wide range of fitness across serotypes. Our findings highlight that all strains are fundamentally different in underlying fitness (Extended Data Fig. 8 ), which means that NVTs will differ in their ability to alter their ecological niche following changes in vaccine formulation. This result needs to be taken into consideration in the development of new vaccine formulations. Among NVTs, serotypes 15A, 35B and 8 had the greatest fitness advantage after PCVs were used (Supplementary Figs. 8 and 9 ), a result concordant with what has been previously observed 20 , 22 , 37 . Shifts in lineage fitness also resulted in changing patterns of spread. By incorporating our fitness estimates for NVT and VT strains after vaccination into our mobility model, we estimated that the number of affected municipalities from a strain of a NVT was 2.02 (95% CI = 1.81–2.25) times the number of affected municipalities from types included in the vaccine (Fig. 3e ).

We next explored whether using the proportion of isolates that were VT versus NVT within each GPSC at the start of the study period and our fitness estimates could explain the subsequent dynamics of individual GPSCs. Simply using VT and NVT fitness estimates could explain 60% of the variance in individual GPSC prevalence at any time. Allowing for serotype-specific differences produced a small improvement, explaining 65% of the variance (Supplementary Fig. 10 and Supplementary Table 13 ). The unexplained variance reflects GPSC-specific fitness and is probably driven by negative-frequency dependent selection (NFDS) 24 . This result highlights the predictive nature of the serotype composition of GPSCs in determining PCV-driven GPSC dynamics (Supplementary Figs. 11 and 12 ).

The serotypes included in the vaccines were prevalent in childhood disease and had high levels of AMR in the USA, where the vaccines were developed 38 . The high levels of AMR in the VT strains was also present globally 26 . In South Africa, similar to other countries, reductions in AMR have been noted since vaccine implementation; however, it remains unclear whether this trend will persist or whether AMR eventually rebounds 26 . In South Africa, before vaccines, 63.6% of VT and 8.8% of NVT strains were resistant to penicillin. We found that there was a clear reduction in overall penicillin resistance following vaccine implementation, which was driven by reductions in the proportions of strains that are VT (Fig. 4f and Supplementary Table 12 ). The trends, although still present, were less clear in the other investigated antimicrobials (Extended Data Fig. 9 ). Owing to the relevance of penicillin as a first-line antimicrobial for pneumococcal disease and the high proportion of resistance in this population, we used our same modelling framework and were able to recover the observed proportions of strains that were resistant to penicillin over time. Before vaccines, among both NVT and VT strains, there was limited difference in the fitness between those that were penicillin-resistant and penicillin-susceptible. However, following implementation of vaccines, NVT-resistant strains were 1.30 (95% CI = 1.19–1.43) times as fit as penicillin-susceptible NVT strains (Fig. 4f,g and Supplementary Fig. 19 ). Conversely, resistance did not seem to have changed among VT penicillin-resistant strains (relative fitness of 0.97 (95% CI = 0.91–1.03)) (Fig. 4f,g , Supplementary Fig. 13A–C and Supplementary Table 12 ). Expansion of NVTs within typically VT-associated lineages is the most common mechanism for serotype replacement 36 . As a result, the penicillin resistance associated with these newly expanded NVT lineages is able to persist in the population 19 , 36 . Together with our quantification of growing penicillin resistance among NVTs following the use of PCVs, this result suggests that the overall reduction in penicillin resistance seen following vaccine implementation may not persist. Our data also highlight the nuanced effect that vaccines can have on patterns of AMR (Supplementary Fig. 13D–F ). It is probable that next-generation higher valency PCVs may also lead to increased AMR prevalence in those serotypes not included in the vaccine. Changing patterns of antimicrobial use in this population may also result in shifts in resistance patterns. The widespread presence of penicillin resistance at the beginning of our data collection period would implicate expansion of resistance among existing lineages; however, we cannot exclude the emergence of some new resistance forms.

Limitations

We did not have complete carriage and invasive disease data across the entire time period. However, we performed sensitivity analyses to determine whether our results are robust to including carriage and invasive disease together. The human mobility data are Meta baseline data that were released by Meta owing to the SARS-CoV-2 pandemic in 2020. We used aggregated data across 17 months (January 2020 to June 2021) within a single mobility pattern matrix. As mobility was altered during this period and to address the possibility that Meta mobility data are different to the movement of individuals involved in pneumococcal transmission, we included an additional parameter that adjusts the human mobility data to account for more or less time being spent in home municipalities. As we were able to obtain good fits to the observed spread of pneumococcus, and these models outperformed standard gravity models, our findings highlight how imperfect mobility data can nevertheless be useful. Vaccination levels for PCV7 were reported to be 89% by 2022 (ref. 39 ). We did not have the data to consider changes in coverage in time or across provinces. Other settings with different levels of coverage may observe different fitness effects from the implementation of the vaccine. Within the fitness model, as we were looking at relative proportions of strains with increasing proportions of resistance, there may still be decreased total burden of resistant disease if those strains carrying it remain low in prevalence.

Here we quantified and explained the movement of a persistent human pathogen for which the geographical course has been largely hidden by its diversity and endemicity. The pneumococcus has an affinity for urban centres through which it channels its wider geographical spread. Although it is characterized by slow transmission overall, the use of vaccines can substantially and rapidly change pneumococcal lineage ecology. Although vaccine-associated fitness dynamics have been previously described in the pneumococcus 8 , 36 , they have not been directly quantified. Increasing proportions of NVTs in the disease isolates from the PCV era can be largely attributed to the decrease in number of VTs rather than the increasing prevalence of NVTs 22 . Vaccination has had a secondary effect on penicillin resistance, with a decrease in recent years in South Africa. Given the estimated growth advantage of penicillin-resistant NVT strains, we may see a reversal of this benefit; however, estimating the carrying capacity of this growth is beyond the scope of this model. Furthermore, we quantified pneumococcal geographical spread and the spatial impact of NVT expansion after vaccination. Our findings highlight how directly observed characterizations of human mobility using mobile phone data or Meta data can be used to obtain a mechanistic understanding of how pathogens spread within phylogeographic frameworks. This includes considering mobility of both infected individuals and the susceptible population and how we can adjust these datasets to account for systematic differences in behaviour between mobile phone data and Meta data and those involved in transmission. We note that basic gravity models also performed well, providing a useful alternative in settings in which mobile phone data options are not available. Our description of pneumococcal geographical spread provides new insight into the movement of emergent strains. In South Africa, the population density in the emergence location of a NVT strain may affect its speed of spread across the country and have implications for public health responses to emergent strains. Emergence in a highly mobile, peri-urban area may enable both rapid proximal geographical spread and less frequent distal seeding events, and thus more complete distribution across the country. The fitness model and human mobility model together provide frameworks to quantify and better understand the migratory and fitness dynamics of this globally endemic pathogen. The magnitude of South Africa and its provinces demonstrates that these frameworks may be applied to other large regions.

Data sources and processing

Pneumococcal sequence data and metadata.

The genomes included in this study were collected as part of the Global Pneumococcal Sequencing project (GPS), which is a global genomic survey of S.   pneumoniae 40 . The invasive disease isolates included here were collected by the National Institute for Communicable Disease in South Africa from 2000 to 2014 (ref. 22 ). In the initial phase of genome sequencing, approximately 300 invasive-disease isolates from each year (2005–2014) were selected with a specific target age breakdown (50% from <3 year olds, 25% from 3–5 year olds, 25% from >5 year olds). In the second phase of sequencing, approximately 200 disease isolates from 2000 to 2004 and approximately 100 invasive disease isolates from 2005 to 2010 from children <5 years old were randomly selected for sequencing 22 (Supplementary Table 2 ). The carriage isolates were collected in Soweto, Gauteng ( n  = 736; collection years 2010, 2012 and 2013) and Agincourt, Mpumalanga ( n  = 1,114; collection years 2009, 2011 and 2013) by the Wits Vaccines and Infectious Diseases Analytics Research Unit. A random sample of 400 carriage isolates from each year were chosen for genome sequencing (Supplementary Table 2 ). We included both carriage and invasive-disease isolates and conducted sensitivity analyses throughout to confirm the methods were robust to both carriage and invasive disease individually. Invasive disease is defined as the bacterium being isolated from a typically sterile site. The majority of total isolates were from children aged <5 years (75.0%); 7.6% were from individuals aged 5–20 years and only 17.5% were from adults (>20 years) (Supplementary Table 7 ). These were distributed across the before and after PCV periods. We additionally included similar GPSCs from the GPS database (for context within the global population) for the RR analysis. We utilized metadata that included collection year and month, residence province of the patient, age of the patient, sampling site and clinical manifestation. The range of sampling sites included nasopharyngeal swabs (for carriage), blood, pleural fluid, cerebrospinal fluid, peritoneal fluid, pus and other joint fluid (for invasive disease).

Population data

We estimated the population for each municipality ( n  = 234) across South Africa using the population-size estimates from LandScan 2017 (refs. 41 , 42 ) (Extended Data Fig. 11 ).

Mobility data

The mobility data used in this study were collected using Meta Data for Good Disaster maps from South Africa. These are initiated at the onset of a disaster—in this case, the SARS-CoV-2 pandemic—and track the geographical movement of Meta users 43 . We used the baseline (adjusting to 2 weeks before) human mobility for each month from January 2020 to July 2021 (refs. 43 , 44 ) to attain a mobility probability from and to each municipality. For each origin (home) municipality ( n  = 234), we determined the mean monthly number of Meta users that were in each destination municipality in South Africa. Location pairs with a value of zero were given a value of ten. We divided each cell by the total number of users across all destinations for that origin municipality. Each cell in the resultant original–destination matrix therefore represented the probability of being in each destination municipality given your home municipality. This is the probability of mobility from each municipality to each other municipality after 2019. Because we do not have mobility data from the exact years the genomes were sampled, we adjusted the diagonal of the mobility matrix using an estimated parameter. This allows people to stay more or less at home and mitigates the effect of non-year matched mobility data. We define the radius of gyration ( R i ) for each municipality as follows:

Where d i is distance to region i , and m i is the probability of mobility to region i . The sum is across all municipalities ( n  = 234) 45 (Extended Data Fig. 11 ).

Generation time distribution

We used a simulation framework to estimate the overall generation time using the separate contributions of the carriage durations and the incubation period. This approach has previously been used for other pathogens 46 .

We sampled 1,000 carriage durations from an exponential distribution with means that are inverse to the clearance rates estimated across serotypes in ref. 31 (clearance rate = 0.026 (95% CI = 0.025–0.028) episode per day) and in ref. 32 (clearance rate = 0.032 (95% CI = 0.030–0.034) episodes per day) 32 .

To sample the day of transmission, we randomly sampled a time point between zero and the time of clearance for each individual. We then separately sampled an incubation period using a uniform distribution of between 1 and 5 days. To account for longer carriages resulting in more opportunities for transmission, we sampled from the distribution of generation, with the probability of sampling each individual weighted by the total carriage duration. The total generation time is then the sum of the duration to transmission and the incubation period.

We repeated these steps 10,000 times and estimated the mean and standard deviation of this final distribution assuming that the generation time follows a gamma distribution with a mean of 35 and a standard deviation of 35 (exponential distribution with a rate of 1/0.096). By comparing the histogram of the distribution with the gamma distribution, this seems a reasonable assumption (Supplementary Fig. 4 ).

Sample culture and genome sequencing

The pneumococcal isolates were selectively cultured on BD Trypticase soy agar II with 5% sheep blood (Beckton Dickinson) and incubated overnight at 37 °C in 5% CO 2 . Genomic DNA was then manually extracted using a modified QIAamp DNA Mini kit (Qiagen) protocol. As part of GPS, pneumococcal isolates were whole-genome sequenced on an Illumina HiSeq platform to produce paired-end reads with an average of 100–125 bp in length, and data were deposited into the European Nucleotide Database. Whole-genome sequence data were processed as previously described 36 , 47 .

We performed predictive antimicrobial susceptibility profiling using the CDC-AMR pipeline for three classes of antimicrobials: β-lactams (penicillin; encoded by the genes pbp1A , pbp2B and pbp2X ) 48 , 49 ; sulfonamides (co-trimoxazole; folA and folP ); and macrolides (erythromycin and clindamycin; ermB and mefA ) 50 , 51 . This was done for 6,798 randomly selected isolates 40 .

Constructing time-resolved phylogenetic trees

We selected the GPSCs for which we had genomes from each of South Africa’s nine provinces and for which we had a minimum of 50 sequences in total to build phylogenies, henceforth referred to as ‘dominant GPSCs’. There were nine dominant GPSCs: GPSC1, GPSC2, GPSC5, GPSC10, GPSC13, GPSC14, GPSC17, GPSC68 and GPSC79 ( n  = 2.575). Assembly was performed using Wellcome Sanger Institute pathogen informatics automated pipelines and is freely available for download from GitHub under an open-source licence, GNU GPL 3 (ref. 52 ). For each sample, sequence reads were used to create multiple assemblies using VelvetOptimiser (v.2.2.5) and Velvet (v.1.2.10) 53 . An assembly improvement step was applied to the assembly with the best N50 and contigs scaffolded using SSPACE (v.2.0) 54 , and sequence gaps were filled using GapFiller (v.1.11) 55 . Assembly quality control parameters included a minimum average sequencing depth of 20× and an assembly length of 1.9–2.3 Mb. Sequences with more than 15% heterozygous SNP sites were excluded.

We created reference genomes for each GPSC using ABACAS (v.1.3.1) to order the contigs from a representative of each GPSC mapped to S.   pneumoniae (strain ATCC 700669/Spain 23F-1) (EMBL accession: FM211187 ) 56 . Any contigs that did not align were concatenated to the end. We multiply mapped all genomes from each dominant GPSC against these references, respectively, using a custom mapping, variant calling and local realignment around indels pipeline (multiple_mappings_to_bam.py) 57 using bwa-MEM (v.0.7.17) 58 and samtools mpileup (v.1.6) 59 . The minimum base quality for a base to be considered was 50. The minimum mapping quality for a SNP to be called was 20, with a minimum of 8 reads matching the SNP. We built trees masking recombination regions using Gubbins (v.2.4.1) 60 with the hybrid model that uses FastTree for the first iteration and RAxML subsequently 61 and a GTR model. We converted branch length to time using BactDating (v.1.0) with a mixed gamma, relaxed clock model 62 . We compared concordance between BEAST (v.1.10.4) 63 with both strict and relaxed clocks, and a Bayesian skyline prior. As the results were concordant, we used BactDating owing to its shorter runtime (Supplementary Fig. 14 ).

RR framework

We used a RR framework to investigate the risk of genetic similarity across geographical distance 64 . We compared the location (loc) and label ( G ) (that is, GPSC or genetic similarity) of pairs of sequences that were collected around the same time ( t ). This approach has been shown to be robust to substantial biases in timing and location of isolate collection 64 . We first constructed pair-wise matrices comparing every isolate to every other isolate ( n pairs = 6,910). In the numerator was the ratio of pairs that were the same GPSC, collected within a year of each other, from the same province, over the total number of pairs collected within a year of each other from the same province. The denominator was the ratio of pairs that were the same GPSC, collected within a year of each other, from distant provinces (>1,000 km apart) (L ref ), over the total number of pairs collected within a year of each other from distant provinces. Geographical distances were calculated based on the centroid coordinates of each province. To demonstrate the suitability of using centroid distances, we simulated a spatial transmission process for 1,000 separate chains in which at each generation, a daughter point is placed at a randomly located location 350 m in each of the x and y direction. This was repeated over 20 generations. We then identified the centroid of each case based on the closest coordinate rounded to the nearest kilometre. We then calculated the total distance covered for both the true distance and the centroid distances and found that the resulting distances were similar (Supplementary Fig. 15 ).

To quantify uncertainty, we used a bootstrapping approach whereby in each bootstrap iteration, we randomly sampled with replacement the isolates before recalculating the statistic. We report the 2.5 and 97.5 percentiles from the resulting distribution.

We also repeated the same analysis but used the time-resolved phylogenetic trees to interrogate pairs across increasing divergence times (breaking the GPSCs into higher resolution). For this, rather than matrices designating whether pairs were the same GPSC or different GPSCs, the divergence time between each pair was included. We only included the divergence times between like GPSCs (Fig. 2c–e ).

We utilized the framework to compare a range of geographical distances, keeping the reference distance to pairs that were >1,000 km apart (Fig. 2 ).

To identify the divergence time at which pairs had an equal risk of being in the same province as distant provinces (time to homogenization across South Africa), we investigated the divergence time at which there was no increased risk of similarity within a province compared against distant provinces (RR = 1). We stratified distances in South Africa into groups of distances, including the 9 within provinces, 14 pairs that were <500 km apart, 16 pairs 500–1,000 km apart and 6 pairs of provinces >1,000 km apart. We repeated the framework across rolling 20-year time windows at 10-year intervals from 0 to 100 years ( g1 , g2 ) (Fig. 2f ). We repeated this for pairs for which one was from South Africa and the other was from another country in Africa (Supplementary Fig. 3A ). We sampled 300 sequences from each province with replacement to compensate for biased sampling (Extended Data Fig. 2a ). Furthermore, to incorporate phylogenetic uncertainty into the statistical framework, we sampled 100 individual phylogenies from the BactDating posterior. We report the 2.5 and 97.5 percentiles from the resulting distribution. We also repeated the analyses only including pairs isolated from patients with pneumococcal disease (Extended Data Fig. 2b ).

Probabilistic mobility model

Overall strategy.

We extended a previously published mechanistic phylogeographic model 65 to estimate the mobility of the pneumococcus between pairs of municipalities at each transmission step. To infer the probable path of transmission between sequence pairs, we used the divergence time and the generation time distribution to estimate the number of transmission generations between pairs of sequences. Each generation is a possible transmission event and provides an opportunity for a mobility event.

The approach ultimately aims to estimate an origin destination matrix for a transmission step, whereby each cell represents the probability that the pneumococcus is now in location j after one transmission step given it was previously in location i . As the phylogenetic trees combined with the generation time provide an estimate of how many transmission steps separate pairs of samples, we can use repeated matrix multiplication to integrate over all possible pathways linking two locations (see below for more details). We incorporated the probability of sampling at each geographical location, within each collection year, by GPSC to account for our observation process.

We follow notation per a previous study 65 . A pair of isolates, C a and C b , with sequences, Seq a and Seq b included in a phylogeny are found in locations, L a and L b , and the samples were taken in the years, T a and T b . The inferred MRCA between C a and C b is time T m and located in L m . The number of transmission generation from the MRCA to C a is G a and to C b is G b .

Model fitting

Single transmission generation.

Considering a single transmission generation, the probability that persons i and j come into contact with each other given i lives in location a and j lives in location b can be written as follows:

Where \(P\left({V}_{i}=k| {L}_{i}=a\right)\) is the probability that individual i , whose home location is in a , visits location k and \(P\left({V}_{j}=k| {L}_{j}=b\right)\) is the probability that individual j whose home location is in b visits location k , and B k is the location-specific probability of transmission for k .

At time \({\tau }\) , one infector in i in location a is expected to transmit to this number of persons in location b :

Where \({S}_{b,{\tau },{\rm{gpsc}}}\) is the number of susceptible people in location b at time \({\tau }\) with some lineage = gpsc.

The total number of people infected by the infector is:

Conditional on transmission occurring, the probability that the infectee’s isolate is taken in location b is:

We then created an NXN transmission matrix, \({\Delta }_{{\tau },{\rm{gpsc}},{\rm{gen}}=1}\) , with N being the total number of locations, containing the transmission probabilities, asymmetrically, between all pairs of locations at each point in time; \({{\delta }}_{a,b,{\tau },{\rm{gpsc}}}\) is element [ a , b ] of the matrix.

Human mobility characterization

We use Meta mobility data (MetMob), as described in the mobility data section above, to characterize mobility between the 234 municipalities of South Africa. We aggregated these to the province level ( n  = 9) to fit the model. Initially we extracted a 234 × 234 matrix that sets out the probability that an individual from municipality a visits location k ; again, where k  = any municipality.

MetMob comes from individuals using Meta; however, this may not be representative of the amount of time spent at home by those involved in pneumococcus transmission. The mean of the diagonal of the Meta Mobility matrix (234 × 234) is 0.989, implying, on average, 98.9% of Meta Users stay in their home municipality, H i . To allow for individuals to spend more or less time at home than represented in the MetMob data, we incorporated a parameter to adjust the probability of being at home ( \({\theta }\) ). We adjusted the probability of staying home using a standard logistic function and restricted it with bounds of –0.04 and 0.6 to facilitate exploration of a sensible space. The adjustment allowed by the bounds limits the range of movement H i –0.6 and H i +0.04.

The probability of a person remaining in location a therefore becomes:

Where a value of greater than 1.0 is obtained for a specific municipality, this is replaced by a value of 0.999.

The estimates thus far are mobility per day, but we were interested in mobility across the infectious period. Therefore, we adjusted the diagonal to account for mobility each day within the infectious period:

We rescaled the probabilities so that the sum of all mobility is equal to 1:

Where MetMob[ a , k ] considers all mobility probabilities from the Meta Mobility data.

The sum of the movements to South African municipalities is equal to 1, which assumes that the analysis contains all possible movements of both the pneumococcus and people and implying no external introductions. The outcome of this is that some mobility may be missed, especially around the country borders.

Probability of the pneumococcus being in each location after G transmission generations

To determine the probability that location k contains the home location after G transmission generations, we used matrix multiplication, which integrates across all possible pathways connecting two locations.

Where t r is the time of generation G r .

Probability of observing a pair of cases in two specific locations

The probability that C A has home location L A and C B has home location L B is conditional on the sequences being observed in locations L A and L B at times T A and T B . We assumed that the location of two cases, L i and L j , is dependent on the location of their MRCA, L m , and the number of transmission generations separating them from their MRCA, G A and G B .

The observations processes across locations are independent of each other, and each transmission event is independent of other transmission events. The probability of observing (Obs) a case at L i at time T A is not dependent on the number of generations to, or location of, the MRCA. We considered discretized space of the nine provinces of South Africa, resulting in the following equation:

Probability of G generations between MRCA and a sequenced isolate

Under the previous equation, \(P({G}_{A},{G}_{B}| {{\rm{Seq}}}_{A},{{\rm{Seq}}}_{B},{T}_{A},{T}_{B})\) represents the generation time distribution.

We can extract the joint probability of C A and C B being separated from the MRCA by G A and G B transmission generations, respectively, using the above-derived generation time distribution and the time-resolved phylogenetic trees.

Assuming the generation time is gamma distributed and all transmission events are independent, the sum of the gamma distribution is also gamma distributed. Additionally, we can extract the evolutionary times E A and E B , separating C A and C B from the MRCA. As previously described 65 , using equation ( 19 ), we can estimate the probability of g transmission events over many trees, allowing us to incorporate phylogenetic including evolutionary parameters from the tree structure.

We determined the probability for the number of generations from MRCA for each isolate for 1–1,000 generations, using the generation time derived above.

Location of the MRCA ( P ( L m ))

We estimated the probability that, on average, an MRCA is in each of the nine provinces in South Africa. We estimated parameters for each of the eight provinces, setting Western Cape aside as a reference, and dividing by the total across all nine to ensure that the sum of the probabilities is 1. This again assumes no external introductions.

Calculation of likelihood

We calculated the likelihood using all pairs of available sequenced S.   pneumoniae as previously described 65 . We accounted for the observation process by incorporating the probability of sampling in each location for isolates belonging to each GPSC annually.

Likelihood equation

We calculated the likelihood using all pairs of sequenced pneumococci as follows:

Where n gpsc are the number of sequences available from GPSC gpsc.

Hamiltonian MCMC

We used an MCMC approach to estimate our parameters using the package fmcmc (v.0.5-1) implemented in R 66 . We estimated nine parameters: a parameter that adjusts the probability of staying in the home municipality compared with Meta mobility data; and eight parameters capturing the relative probability that the MRCA of a pair of individuals was in each of the other eight provinces compared with the province of Western Cape (the reference).

We only used pairs of sequences that were separated by less than 10 years of evolutionary time between them. After 10 years, there are limited spatial signals remaining, as the bacterium has had many opportunities to move. This approach was used to make the model computationally tractable. To incorporate phylogenetic uncertainty, we repeatedly refit the model using 50 randomly selected phylogenies from the BactDating posterior. We report the 2.5 and 97.5 percentiles from the resulting distribution. In a sensitivity analysis, we showed that increasing the model to 15 years resulted in similar estimates (Extended Data Fig. 12a ).

Model performance

To assess the performance of our model, we used a simulation framework. We simulated 50,000 pairs of events, locations and the location of the MRCA between pairs, whereby the probability of mobility between each pair of locations was determined by the Meta mobility matrix adjusted by a parameter of known value of −2. We tested our ability to recapture this input parameter. To incorporate the biased observation process, we downsampled the simulated data based on the by-province sampling probabilities from our true data.

We used the downsampled data to fit our model with 20,000 steps of a MCMC with a jump step of 0.08 in 3 chains. We were able to recapture the downsampled data utilizing the human mobility framework and the estimated parameter for the probability of staying at home, accounting for the sampling probability per province. We then utilized the same human mobility framework and the estimated parameter, excluding the sampling probability, and were able to recapture the complete simulated data, including the input parameter (Extended Data Fig. 5a,b ).

We repeated these simulations but downsampled with various biases. We determined how well the model performed when we only sampled two provinces. We also tested how far off our estimates would be if our generation time estimate was 50% higher or 50% lower than we had estimated given the Kenyan and Gambian data 31 , 32 (Extended Data Fig. 5c,d ).

Model sensitivity analyses

We estimated our parameters only to include isolates from patients with invasive pneumococcal disease (Supplementary Fig. 1B ).

To test the impact of a range of generation times on the parameter estimates, we also re-ran our MCMC with generation times of 15, 35 (as included in the model) and 55 days. To quantify uncertainty, we sampled posterior parameters, reporting the 2.5 and 97% percentiles (Extended Data Fig. 12b ).

We confirmed that the chains converged for each of the nine parameters estimated (Supplementary Fig. 16 ).

Probability guided transmission simulations

Person-to-person transmission chains.

We simulated person-to-person transmission chains seeded in a starting municipality weighted by the population size. We determined the RR of being in a specific municipality after 10 transmission generations (approximately 1 year) across 500,000 simulations.

We fixed the starting municipality to be rural (population density <50 km –2 ) or urban (population density >500 km –2 ) and repeated the above simulation. For 10,000 sequential simulations, we counted the number of unique municipalities affected and distance travelled at each transmission interval weighting by population size. We then determined the number of municipalities travelled to across all transmission chains (Extended Data Fig. 3 ).

Branching epidemic

We simulated a branching epidemic in which we drew the number of transmission events seeded by each event from a Poisson distribution around an effective reproductive number ( R eff ) of 1 and amplitude of 0.15 over 100 iterations. We determined the mean distance from the starting municipality after 60 generations (5.8 years with a 35-day generation time) and the number of municipalities visited over that time. We calculated the uncertainty at the 95% CI of a normal distribution.

Gravity model

We compared the performance of the model that used Meta data with the performance of a simple gravity model whereby the probability of mobility is determined by the distance and human population size of the municipalities. We calculated the probability of mobility between locations i and j (GravMob i,j ) to be the log of the destination population size (popsize j ) raised to parameter β divided by the distance between locations, loc i and loc j (dist i,j ) raised to parameter γ .

We also tested a model including only distance and estimating an exponent adjustment parameter γ .

We calculated the DIC comparing the three models and found that the Meta mobility model (DIC = 12,290.97) was the best-performing model 67 when compared with the gravity model (DIC = 12,294.34). Both of these models performed better than distance alone (DIC = 12,424.32) (Extended Data Fig. 4 ).

All statistical analysis for the RR framework and the mobility model was performed in R (v.3.6.2) 68 .

Population-level fitness model

We developed logistic growth models to fit the changing prevalence of different serotypes and assess the impact of vaccine implementation, utilizing a method that has previously been implemented for the endemic bacterium Bordetella pertussis 69 .

Vaccine-type model

We first binned serotypes into three groups: those not included in the vaccine (NVTs), serotypes included in PCV7 (4, 6B, 9V, 14, 18C, 19F and 23F), and additional serotypes included in PCV13 (1, 3, 5, 6A, 7F and 19A). We used the full data for this analysis. We computed the relative abundance f i ,ref of each group of serotypes, i , compared with a reference type, ref.

We chose to use NVTs as the reference. Varying the reference did not affect the model, as long as the reference samples span all years. We then used a simple logistic model, assuming a constant growth rate to capture the evolution of this abundance, at each time t .

where r i ,ref is the growth rate of that abundance shared across all provinces, and f i ,ref,0 is the initial relative abundance of the serotype group i with respect to a chosen ref.

To control for the varying presence of all circulating serotypes through time, we present fitness as the average relative growth rate, \(\bar{{r}_{i}}\) , for each group with respect to a randomly selected group in the population:

where n is the number of groups, and \(\bar{{f}_{j}}\) is the average absolute frequency of the group in the period of time considered.

This average relative growth rate, \(\bar{{r}_{i}}\) , can be identified as the selection rate coefficient of the group in the population considered 69 . The selection rate coefficient is a direct measure of the fitness advantage of emerging variants and is one of the best indicators as to whether a strain will increase in frequency during an outbreak 70 , 71 .

We can further multiply the selection rate by the mean generation time (35 days) to obtain the selection coefficient per generation, which is the relative fitness advantage per transmission generation.

In the Article, we also present estimates of the relative fitness advantage of groups in particular time frames. We used the same computation as for the average relative growth rate, but tailored it to specific references. For example, to compute the relative fitness advantage of NVTs compared with VTs in the PCV era \(\Delta \overline{\,{r}_{{\rm{NVT}},{\rm{VT}},{\rm{after}}{\rm{PCV}}}}\) , we computed:

To fit the model, we used Hamiltonian Monte Carlo as implemented in R-Stan 72 , with stan (v.2.26.1). We used a Poisson likelihood to fit the observed proportion of sequences that were of each category and the total number of isolates in that year as an offset. The model estimated the proportion of isolates that were of each type at the start of the dataset and the fitness parameters.

Exploration of vaccine effect

To investigate whether the serotypes fitness changed across after implementation of PCV7 and PCV13, we tested a range of models. We considered a model without any shift of fitness, a model with a single shift in 2009 and a model with two growth rates represented by a shift in fitness in both 2009 and 2011. The 2009 shift pertains only to those serotypes included in PCV7 and the 2011 shift pertains to those additional 6 serotypes in PCV13. Model comparison was done using the Watanabe–Akaike information criterion (WAIC) implemented in the loo package 73 . Furthermore, we tested for a potential delay between implementation of PCVs and the change in fitness. Model comparison is presented in Extended Data Fig. 7 .

The model performed best with an initial fitness switch in 2009 (the initial year of vaccine implementation) and another fitness switch in 2011 (the year of PCV13 implementation). We used this model in the main text.

Estimation of individual serotype fitness

Using the same framework, we also estimated the growth rate per serotype to capture whether the individual dynamics were concordant with, or deviated from, what was expected according to their respective group (NVT, PCV7 or PCV13) (Fig. 4a–c ). The reference strain in this analysis was set to serotype 13, which is a NVT and for which samples span all years.

Exploration of predicted GPSC dynamics based on their serotype composition

We explored whether using the proportion of isolates that were VT versus NVT within each GPSC at the start of the study period and our fitness estimates could explain the subsequent dynamics of individual GPSCs.

To predict the dynamics of individual GPSCs based on their serotype composition, we used the same framework as described above, applied to each GPSC serotype group in our dataset. As the number of such groups is large ( n  = 340 GPSC serotype groups), we restricted the analysis to the GPSCs present at a minimum prevalence of 1%. This led to 26 GPSCs being considered for this analysis (representing 74.7% of the dataset), split in a total of 101 GPSC serotype groups. We then modelled the dynamics of each GPSC serotype group, estimating one starting frequency per group and using the previously estimated fitness parameters, either by VT or serotype. We also considered a model with no fitness parameters (relative fitness = 1). To assess the model performances, we computed the Akaike information criterion (AIC) 74 using the average likelihood of each model and the number of parameters used in each model. We also computed the predicted GPSC dynamics over time for each model by summing all the predicted GPSC serotype group dynamics for each GPSC. As a measure of goodness of fit we used the coefficient of determination of the observed versus predicted GPSC proportions each year:

R 2 is also the proportion of variation explained by the variables considered in each model.

Our model estimated a constant fitness for each group considered (VTs or individual serotypes), assuming that if a group has the highest fitness, it will eventually replace the other groups. This assumption is meaningful for VTs and serotypes, as some are directly targeted by the vaccines, a strong selective force in the population. However, the fitness of each GPSC cannot be modelled with this simple assumption as it has been shown that their fitness is inherently multifactorial, as described in the NFDS model 24 .

We then used the model to capture the decreasing proportion of penicillin resistance in the population. To do this, we incorporated the dynamics of VTs (PCV7 and PCV13) and NVTs in the population and the respective proportion of penicillin resistance within them. To keep the model tractable, we did not differentiate here between serotypes included in PCV7 or PCV13, instead we group them into VTs. We use the PCV7 implementation (2009) as the year of fitness shift. This parametrization marginally differed from the best model (next best) (Extended Data Fig. 7 ) and enabled us to keep the number of categories tractable. We also compared the single switch model to a model with no change in fitness at the time of vaccine implementation. This approach showed that the model with a switch was superior (Supplementary Fig. 17 ).

We used the same approach described above to model the proportion of VTs, f VT , and NVTs, f NVT . We then modelled the proportion of strains that were and were not penicillin resistant within each group, either VT, f AMR|VT or NVT, f AMR|NVT . We estimated the fitness of resistance in each.

We then fit the model to all four groups, f VX,AMR (resistant NVT, susceptible NVT, resistant VT and susceptible VT).

Where f VX is the proportion of either VT or NVT in the population and f AMR|VX is the proportion of penicillin resistance within each of those groups. We then derived the proportion of penicillin resistance overall in the population by summing the proportion of VT resistant, f VT,AMR , and NVT resistant, f NVT,AMR , strains.

Estimating effect of fitness on migration

We included the calculated average relative growth rates in the simulated branching epidemic to calculate the mean number of municipalities visited, distance travelled and probability of being in the home municipality over 5 years. We report uncertainty at one standard deviation from the mean. We repeated this incorporating the post-PCV selection coefficients for NVTs and VTs and estimated the relative increase in the number of municipalities visited for NVT serotypes compared with PCV serotypes after 2 years of transmission.

Fitness model carrying capacity sensitivity

Our fitness model assumed constant fitness over time. This specifically means that if a serotype is found to be fitter than the rest of the population, we expect it to replace the whole population after some time. However, the currently best-supported model proposes that NFDS drives pneumococcus population dynamics 24 , whereby lower frequency genes become more fit.

To test the effect of our assumption on our fitness estimates, we performed a sensitivity analysis. We introduced minimum and maximum carrying capacities in our logistic model. In equation ( 20 ), we let the relative frequency f i ,ref go to a maximum of K max and in equation ( 21 ), a minimum of K min :

with \({K}_{\min }\le {K}_{\max }\) , and \({K}_{\min }\in \left[0,1\right],\,{K}_{\max }\in \left[0,1\right]\) .

We considered a range of values for K min (0.1, 0.05 and 0) and K max (0.9, 0.95 and 1) (Supplementary Fig. 20 ). We tested the effect of this carrying capacity has on both the vaccine status model (Supplementary Fig. 18 ) and the AMR and vaccine status model (Supplementary Fig. 19 ). In each case, we compared the fitness estimates obtained. We found that the fitness estimates were robust to variable carrying capacities across those tested, which implied that our assumption does not affect this framework. However, it is important to note that this model is robust to set variable carrying capacities while being unable to estimate them, which is in contrast to NFDS. Our model assumes that there is a population replacement up to the carrying capacity and does not allow for fine-scale estimates of an equilibrium. This fitness model can be used as a quantitative descriptor of the growth of distinct groups (that is, penicillin resistance or VT serotypes in the context of a perturbation) but cannot describe the complex underlying fitness effects of changing frequencies of genes under NFDS 24 .

All statistical analysis for the fitness model was performed in R (v.4.0.5) 68 .

The study was coordinated by the GPS ( https://www.pneumogen.net/gps/ ), whose activities are approved by respective ethics committee in-country. Isolates sequenced in this study come from the National Institute for Communicable Diseases and the University of Witwatersrand carriage studies as previously published. These isolates were all collected as part of existing public health surveillance approved protocols in each country. No personally identifiable information was used as part of this study.

Reporting summary

Further information on research design is available in the  Nature Portfolio Reporting Summary linked to this article.

Data availability

All data and code for figures and analysis are accessible at GitHub ( https://github.com/sophbel/geomig_evo_pneumo ). All whole-genome sequences were deposited into the European Nucleotide Database and accession numbers are available in the GitHub repository and on FigShare ( https://doi.org/10.6084/m9.figshare.24219214 ) 75 . Associated metadata are available from the Microreact webserver ( https://microreact.org/project/7wqgd2gbBBEeBLLPKonbaT-belman2024southafricapneumococcus ), as well as from the GitHub repository and Global Pneumococcal Sequencing Project Monocle database ( https://data.monocle.sanger.ac.uk/ ).

Code availability

All code and scripts for analysis and figures are available from GitHub ( https://github.com/sophbel/geomig_evo_pneumo ).

World Health Organization. The top 10 causes of death. WHO https://www.who.int/news-room/fact-sheets/detail/the-top-10-causes-of-death (2020).

Troeger, C. et al. Estimates of the global, regional, and national morbidity, mortality, and aetiologies of lower respiratory tract infections in 195 countries: a systematic analysis for the Global Burden of Disease Study 2015. Lancet Infect. Dis. 17 , 1133–1161 (2017).

Article   Google Scholar  

Ikuta, K. S. et al. Global mortality associated with 33 bacterial pathogens in 2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 400 , 2221–2248 (2022).

Bender, R. G. et al. Global, regional, and national incidence and mortality burden of non-COVID-19 lower respiratory infections and aetiologies, 1990–2021: a systematic analysis from the Global Burden of Disease Study 2021. Lancet Infect. Dis. https://doi.org/10.1016/S1473-3099(24)00176-2 (2024).

Lees, J. A. et al. Fast and flexible bacterial genomic epidemiology with PopPUNK. Genome Res. 29 , 304–316 (2019).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Ganaie, F. et al. A new pneumococcal capsule type, 10D, is the 100th serotype and has a large cps fragment from an oral Streptococcus . mBio 11 , e00937-20 (2020).

Article   PubMed   PubMed Central   Google Scholar  

Ganaie, F. A. et al. Discovery and characterization of pneumococcal serogroup 36 capsule subtypes, serotypes 36 A and 36B. J. Clin. Microbiol. 61 , e0002423 (2023).

Article   PubMed   Google Scholar  

GPS Consortium, Global Pneumococcal Sequencing collection. Microb. Genomics https://www.microbiologyresearch.org/content/global-pneumococcal-sequencing-collection (2022).

Adegbola, R. A. et al. Carriage of Streptococcus pneumoniae and other respiratory bacterial pathogens in low and lower-middle income countries: a systematic review and meta-analysis. PLoS ONE 9 , e103293 (2014).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Wahl, B. et al. Burden of Streptococcus pneumoniae and Haemophilus influenzae type b disease in children in the era of conjugate vaccines: global, regional, and national estimates for 2000-15. Lancet Glob. Health 6 , e744–e757 (2018).

Jacobs, M. R. et al. Emergence of multiply resistant pneumococci. N. Engl. J. Med. 299 , 735–740 (1978).

Article   CAS   PubMed   Google Scholar  

Hume-Nixon, M. et al. Systematic review of the clinical outcomes of pneumonia with a penicillin-group resistant pneumococcus in respiratory and blood culture specimens in children in low- and middle-income countries. J. Glob. Health 12 , 10004 (2022).

Murray, C. J. et al. Global burden of bacterial antimicrobial resistance in 2019: a systematic analysis. Lancet 399 , 629–655 (2022).

Article   CAS   Google Scholar  

Musher, D. M., Anderson, R. & Feldman, C. The remarkable history of pneumococcal vaccination: an ongoing challenge. Pneumonia 14 , 5 (2022).

Pfizer. Prevenar 13. Pfizer https://www.pfizer.co.uk/products/prescription-medicines/prevenar-13 (2023).

GlaxoSmithKline. Synflorix TM , GlaxoSmithKline’s pneumococcal vaccine, receives European authorisation. GSK https://www.gsk.com/en-gb/media/press-releases/synflorix-glaxosmithkline-s-pneumococcal-vaccine-receives-european-authorisation/ (2023).

Kobayashi, M. et al. Use of 15-valent pneumococcal conjugate vaccine and 20-valent pneumococcal conjugate vaccine among U.S. adults: updated recommendations of the Advisory Committee on Immunization Practices—United States, 2022. MMWR Morb. Mortal. Wkly Rep. 71 , 109–117 (2022).

VIEW-hub (Johns Hopkins Bloomberg School of Public Health, International Vaccine Access Center (IVAC), www.view-hub.org (2022).

Ndlangisa, K. M. et al. The Global Pneumococcal Sequencing ConsortiumYR 2022, a Streptococcus pneumoniae lineage usually associated with pneumococcal conjugate vaccine (PCV) serotypes is the most common cause of serotype 35B invasive disease in South Africa, following routine use of PCV. Microb. Genom. 8 , 000746 (2022).

CAS   PubMed   PubMed Central   Google Scholar  

Javaid, N. et al. Population genomics of pneumococcal carriage in South Africa following the introduction of the 13-valent pneumococcal conjugate vaccine (PCV13) immunization. Microb. Genom. 8 , 000831 (2022).

CAS   Google Scholar  

Belman, S. et al. Genetic background of Cambodian pneumococcal carriage isolates following pneumococcal conjugate vaccine 13. Microb. Genom. 8 , mgen000837 (2022).

PubMed   PubMed Central   Google Scholar  

von Gottberg, A. et al. GERMS-SA Investigators, effects of vaccination on invasive pneumococcal disease in South Africa. N. Engl. J. Med. 371 , 1889–1899 (2014).

Huang, L. et al. Cost-effectiveness analysis of the South African Infant National Immunization Program for the prevention of pneumococcal disease. Infect. Dis. Ther. 12 , 933–950 (2023).

Corander, J. et al. Frequency-dependent selection in vaccine-associated pneumococcal population dynamics. Nat. Ecol. Evol. 1 , 1950–1960 (2017).

Azarian, T. et al. Frequency-dependent selection can forecast evolution in Streptococcus pneumoniae . PLoS Biol. 18 , e3000878 (2020).

Andrejko, K., Ratnasiri, B., Hausdorff, W. P., Laxminarayan, R. & Lewnard, J. A. Antimicrobial resistance in paediatric Streptococcus pneumoniae isolates amid global implementation of pneumococcal conjugate vaccines: a systematic review and meta-regression analysis. Lancet Microbe 2 , e450–e460 (2021).

Müller, N. F., Dudas, G. & Stadler, T. Inferring time-dependent migration and coalescence patterns from genetic sequence and predictor data in structured populations. Virus Evol. 5 , vez030 (2019).

Skosana, Z. et al. Non-vaccine serotype pneumococcal carriage in healthy infants in South Africa following introduction of the 13-valent pneumococcal conjugate vaccine. S. Afr. Med. J. 111 , 143–148 (2021).

Madhi, S. A. et al. Residual colonization by vaccine serotypes in rural South Africa four years following initiation of pneumococcal conjugate vaccine immunization. Expert Rev. Vaccines 19 , 383–393 (2020).

Movement Range Maps , Facebook Data for Good , https://dataforgood.fb.com/tools/movement-range-maps/ (Meta, 2022).

Chaguza, C. et al. Carriage dynamics of pneumococcal serotypes in naturally colonized infants in a rural African setting during the first year of life. Front. Pediatr. 8 , 587730 (2021).

Abdullahi, O. et al. Rates of acquisition and clearance of pneumococcal serotypes in the nasopharynges of children in Kilifi District, Kenya. J. Infect. Dis. 206 , 1020–1029 (2012).

Ginsburg, C. et al. Human capital on the move: education as a determinant of internal migration in selected INDEPTH surveillance populations in Africa. Demogr. Res. 34 , 845–884 (2016).

Collinson, M. A. Striving against adversity: the dynamics of migration, health and poverty in rural South Africa. Glob. Health Action https://doi.org/10.3402/gha.v3i0.5080 (2010).

Kleynhans, J. et al. Estimated impact of the pneumococcal conjugate vaccine on pneumonia mortality in South Africa, 1999 through 2016: an ecological modelling study. PLoS Med. 18 , e1003537 (2021).

Gladstone, R. A. et al. International genomic definition of pneumococcal lineages, to contextualise disease, antibiotic resistance and vaccine impact. EBioMedicine 43 , 338–346 (2019).

Müller, A. et al. Streptococcus pneumoniae serotypes associated with death, South Africa, 2012–2018. Emerg. Infect. Dis. 28 , 166–179 (2022).

US Centers for Disease Control and Prevention. Pneumococcal Disease and Antibiotic Resistance. CDC https://www.cdc.gov/pneumococcal/php/drug-resistance/index.html (2022).

VIEW-hub—South Africa (Johns Hopkins Bloomberg School of Public Health, International Vaccine Access Center (IVAC)), https://view-hub.org/map/country/zaf (accessed September 2022).

GPS, L. McGee, S. D. Bentley, K. P. Klugman. GPS: Global Pneumococcal Sequencing Project , https://www.pneumogen.net/gps/ (GPS Project, 2022).

Mesev, V. in Remotely-Sensed Cities (ed. Mesev, V.) 301–314 (CRC Press, 2003).

ORNL LandScan Viewer , https://landscan.ornl.gov/ (Oak Ridge National Laboratory, 2022).

Maas, P. et al. Facebook Disaster Maps: Methodology. Meta Research https://research.facebook.com/blog/2017/6/facebook-disaster-maps-methodology/ (2017).

Maas, P. Facebook disaster maps: aggregate insights for crisis response & recovery. In Proc. 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (eds Teredesai, A. & Kumar. V.) 3173 (ACM, 2019).

Wesolowski, A. et al. Quantifying the impact of accessibility on preventive healthcare in Sub-Saharan Africa using mobile phone data. Epidemiology 26 , 223–228 (2015).

Salje, H. et al. Reconstruction of 60 years of chikungunya epidemiology in the Philippines demonstrates episodic and focal transmission. J. Infect. Dis. 213 , 604–610 (2016).

Page, A. J. et al. Robust high-throughput prokaryote de novo assembly and improvement pipeline for Illumina data. Microb. Genom. 2 , e000083 (2016).

Li, Y. et al. Penicillin-binding protein transpeptidase signatures for tracking and predicting β-lactam resistance levels in Streptococcus pneumoniae . mBio 7 , e00756-16 (2016).

Li, Y. et al. Validation of β-lactam minimum inhibitory concentration predictions for pneumococcal isolates with newly encountered penicillin binding protein (PBP) sequences. BMC Genomics 18 , 621 (2017).

Metcalf, B. J. et al. Using whole genome sequencing to identify resistance determinants and predict antimicrobial resistance phenotypes for year 2015 invasive pneumococcal disease isolates recovered in the United States. Clin. Microbiol. Infect. 22 , 1002.e1–1002.e8 (2016).

Metcalf, B. J. BenJamesMetcalf/Sanger_SPN_Scripts-Ref. GitHub https://github.com/BenJamesMetcalf/Sanger_SPN_Scripts-Ref (2016).

Sanger Pathogen Informatics. sanger-pathogens/vr-codebase. GitHub https://github.com/sanger-pathogens/vr-codebase (2022).

Zerbino, D. R. & Birney, E. Velvet: algorithms for de novo short read assembly using de Bruijn graphs. Genome Res. 18 , 821–829 (2008).

Boetzer, M., Henkel, C. V., Jansen, H. J., Butler, D. & Pirovano, W. Scaffolding pre-assembled contigs using SSPACE. Bioinformatics 27 , 578–579 (2011).

Boetzer, M. & Pirovano, W. Toward almost closed genomes with GapFiller. Genome Biol. 13 , R56 (2012).

Assefa, S., Keane, T. M., Otto, T. D., Newbold, C. & Berriman, M. ABACAS: algorithm-based automatic contiguation of assembled sequences. Bioinformatics 25 , 1968–1969 (2009).

Wellcome Sanger Institute. sanger-pathogens/bact-gen-scripts. GitHub https://github.com/sanger-pathogens/bact-gen-scripts (2023).

Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows–Wheeler transform. Bioinformatics 25 , 1754–1760 (2009).

Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25 , 2078–2079 (2009).

Croucher, N. J. et al. Rapid phylogenetic analysis of large samples of recombinant bacterial whole genome sequences using Gubbins. Nucleic Acids Res. 43 , e15 (2015).

Stamatakis, A. RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogenies. Bioinformatics 30 , 1312–1313 (2014).

Didelot, X., Croucher, N. J., Bentley, S. D., Harris, S. R. & Wilson, D. J. Bayesian inference of ancestral dates on bacterial phylogenetic trees. Nucleic Acids Res. 46 , e134 (2018).

Drummond, A. J. & Bouckaert, R. R. Bayesian Evolutionary Analysis with BEAST (Cambridge Univ. Press, 2015).

Salje, H. et al. Dengue diversity across spatial and temporal scales: local structure and the effect of host population size. Science 355 , 1302–1306 (2017).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Salje, H. et al. Reconstructing unseen transmission events to infer dengue dynamics from viral sequences. Nat. Commun. 12 , 1810 (2021).

Yon, G. V. & Marjoram, P. fmcmc: a friendly MCMC framework. R package version 0.5-1 https://CRAN.R-project.org/package=fmcmc (2022).

Spiegelhalter, D. J., Best, N. G., Carlin, B. P. & Van Der Linde, A. Bayesian measures of model complexity and fit. J. R. Stat. Soc. Series B Stat. Methodol. 64 , 583–639 (2002).

Article   MathSciNet   Google Scholar  

The R Developmennt Core Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2022).

Lefrancq, N. et al. Global spatial dynamics and vaccine-induced fitness changes of Bordetella pertussis . Sci. Transl. Med. 14 , eabn3253 (2022).

Day, T. & Gandon, S. Applying population-genetic models in theoretical evolutionary epidemiology. Ecol. Lett. 10 , 876–888 (2007).

Davies, N. G. et al. Estimated transmissibility and impact of SARS-CoV-2 lineage B.1.1.7 in England. Science 372 , eabg3055 (2021).

Goodrich, B., Gabry, J., Ali, I. & Brilleman, S. rstanarm: Bayesian applied regression modeling via Stan. R package version 2.21.1 https://mc-stan.org/rstanarm/ (2020).

Vehtari, A. et al. loo: efficient leave-one-out cross validation and WAIC for Bayesian models v.0.1.0, https://doi.org/10.32614/CRAN.package.loo (2015).

Akaike, H. in Selected Papers of Hirotugu Akaike (eds Parzen, E. et al.) 199–213 (Springer, 1998).

Belman, S. Accession Numbers. FigShare https://doi.org/10.6084/m9.figshare.24219214 (2024).

Download references

Acknowledgements

This work was supported by the Wellcome Trust (grant number WT QQ2016-2021, reference 206194, to S.B., S.L. and S.D.B.); the Bill & Melinda Gates Foundation (under Investment ID INV-003570 to S.A.M., A.v.G., M.d.P., S.D. and S.N.); the NIH (grant number R01AI160780 to H.S.); and the European Research Council (grant number 804744 to H.S. and N.L.). For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. We would like to thank all Global Pneumococcal Sequencing project partners; and M. O’Driscoll, S. Farr, and V. Carr for code review.

Author information

These authors jointly supervised this work: Stephen D. Bentley, Henrik Salje

Authors and Affiliations

Parasites and Microbes, Wellcome Sanger Institute, Hinxton, UK

Sophie Belman, Stephanie W. Lo, Kate Mellor & Stephen D. Bentley

Department of Genetics, University of Cambridge, Cambridge, UK

Sophie Belman, Noémie Lefrancq & Henrik Salje

Global Health Resilience, Earth Sciences Department, Barcelona Supercomputing Center, Barcelona, Spain

Sophie Belman

Division of Public Health Surveillance and Response, National Institute for Communicable Diseases of the National Health Laboratory Service, Johannesburg, South Africa

Susan Nzenze

South African Medical Research Council Vaccines and Infectious Diseases Analytics Research Unit, School of Pathology, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa

Sarah Downs & Shabir A. Madhi

Department of Clinical Microbiology and Infectious Diseases, School of Pathology, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa

Mignon du Plessis & Anne von Gottberg

Centre for Respiratory Diseases and Meningitis, National Institute for Communicable Diseases of the National Health Laboratory Service, Johannesburg, South Africa

Mignon du Plessis, Kedibone Ndlangisa, Linda De Gouveia, Mushal Ali, Nicole Wolter, Cebile Lekhuleni & Anne von Gottberg

Milner Centre for Evolution, Department of Life Sciences, University of Bath, Bath, UK

Stephanie W. Lo

National Center for Immunization and Respiratory Diseases, Centers for Disease Control and Prevention, Atlanta, GA, USA

Bernard Beall, Paulina A. Hawkins & Lesley McGee

Department of Science and Technology/National Research Foundation, South African Research Chair Initiative in Vaccine Preventable Diseases, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa

Shabir A. Madhi

Division of Medical Microbiology, Department of Pathology, Faculty of Health Sciences, University of Cape Town, Cape Town, South Africa

Anne von Gottberg

Servicio Antimicrobianos, National Reference Laboratory (NRL), Instituto Nacional de Enfermedades Infecciosas (INEI)-ANLIS ‘Dr Carlos G. Malbrán’, Buenos Aires, Argentina

Alejandra Corso & Paula Gagetti

International Centre for Diarrheal Diseases Research, Dhaka, Bangladesh

Abdullah W. Brooks

Child Health Research Foundation, Dhaka, Bangladesh

Md Hasanuzzaman, Samir K. Saha & Senjuti Saha

Department for Microbiology, Virology and Immunology, Belarusian State Medical University, Minsk, Belarus

Alexander Davydov & Leonid Titov

Center of Bacteriology, Institute Adolfo Lutz, São Paulo, Brazil

Samanta Cristine Grassi Almeida

Cambodia–Oxford Medical Research Unit, Angkor Hospital for Children, Siem Reap, Cambodia

Paul Turner

Peking University People’s Hospital, Beijing, China

Chunjiang Zhao & Hui Wang

Department of Microbiology, Faculty of Medicine, The Chinese University of Hong Kong–Prince of Wales Hospital, Hong Kong, China

Margaret Ip

Department of Microbiology and Carol Yu Centre for Infection, The University of Hong Kong–Queen Mary Hospital, Hong Kong, China

Pak Leung Ho & Pierra Law

Francis I. Proctor Foundation and Department of Ophthalmology, University of California San Francisco, San Francisco, CA, USA

Jeremy D. Keenan

Clinical Research Center, Centre Hospitalier Intercommunal de Créteil, Créteil, France

Robert Cohen

Université Paris Est, IMRB-GRC GEMINI, Créteil, France

Groupe de Pathologie Infectieuse Pédiatrique de la Société Française de Pédiatrie, Nice, France

National Reference Center for Pneumococci, Data and Research Department – DRIM, Centre Hospitalier Intercommunal de Créteil, Créteil, France

Emmanuelle Varon

Department of Medical Microbiology, University of Ghana Medical School, Accra, Ghana

Eric Sampane-Donkor

Department of Clinical Microbiology, Christian Medical College, Vellore, India

Balaji Veeraraghavan

Central Research Laboratory, Kempegowda Institute of Medical Sciences, Bangalore, India

Geetha Nagaraj, K. L. Ravikumar, J. Yuvaraj & Varun Shamanna Noga

The Shraga Segal Department of Microbiology, Immunology and Genetics, Faculty of Health Sciences, Ben-Gurion University of the Negev, Beer-Sheva, Israel

Rachel Benisty & Ron Dagan

Center for Global Health Research, Kenya Medical Research Institute (KEMRI), Kisumu, Kenya

Godfrey Bigogo

Division of Bacterial Diseases, National Center for Immunization and Respiratory Diseases, Centers for Disease Control and Prevention, Atlanta, GA, USA

Jennifer Verani

Malawi–Liverpool–Wellcome Trust Clinical Research Programme, University of Malawi College of Medicine, Blantyre, Malawi

Anmol Kiran, Jennifer Cornick, Maaike Alaerts, Brenda Kwambana-Adams, Ebenezer Foster-Nyarko, Ebrima Bojang, Martin Antonio & Peggy-Estelle Tientcheu

Department of Public Health and Epidemiology, College of Medicine and Health Sciences, Khalifa University, Abu Dhabi, UAE

Dean B. Everett

Infection Research Unit, Khalifa University, Abu Dhabi, UAE

Faculty of Applied Sciences, UCSI University, Kuala Lumpur, Malaysia

Shamala Devi Sekaran

Faculty of Medicine and Institute for Life Sciences, University of Southampton, Southampton, UK

Stuart C. Clarke

Centro de Investigação em Saúde da Manhiça, Maputo, Moçambique

Benild Moiane, Betuel Sigauque & Helio Mucavele

Oxford Vaccine Group, Department of Paediatrics, University of Oxford, NIHR Oxford Biomedical Research Centre, Oxford, UK

Andrew J. Pollard

Faculty of Medicine and Health, University of Sydney, Sydney, New South Wales, Australia

Rama Kandasamy

Institute of Environmental Science and Research Limited, Kenepuru Science Centre, Porirua, New Zealand

Philip E. Carter

Division of Pediatric Infectious Diseases, University of Alabama at Birmingham (UAB), Birmingham, AL, USA

Stephen K. Obaro

International Foundation against Infectious Diseases in Nigeria (IFAIN), Abuja, Nigeria

Wesfarmers Centre of Vaccines and Infectious Diseases, Telethon Kids Institute, The University of Western Australia, Nedlands, Western Australia, Australia

Deborah Lehmann

Infection and Immunity Unit, Papua New Guinea Institute of Medical Research, Goroka, Papua New Guinea

Rebecca Ford

Instituto de Medicina Tropical Alexander von Humboldt, Universidad Peruana Cayetano Heredia, Lima, Peru

Theresa J. Ochoa

National Medicines Institute, Warsaw, Poland

Anna Skoczynska, Ewa Sadowy, Waleria Hryniewicz & Weronika Puzia

Hamad Medical Corporation, Doha, Qatar

Sanjay Doiphode

G. N. Gabrichevsky Research Institute for Epidemiology and Microbiology, Moscow, Russia

Ekaterina Egorova, Elena Voropaeva & Yulia Urban

Department for Public Health Microbiology, National Laboratory of Health, Environment and Food, Ljubljana, Slovenia

Tamara Kastrin

Department of RDI Microbiology, Institut de Recerca Sant Joan de Deu, Hospital Sant Joan de Deu, Universitat Internacional de Catalunya and CIBER Epidemiology and Public Health (CIBERESP), Barcelona, Spain

Carmen Muñoz Almagro, Alba Redin Alonso & Desiree Henares

Faculty of Pharmacy, Siam University, Bangkok, Thailand

Somporn Srifuengfung

Centre for Epidemic Preparedness and Response, London School of Hygiene and Tropical Medicine, London, UK

Martin Antonio

Department of Infection Biology, Faculty of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, London, UK

Medical Research Council Unit The Gambia at London School of Hygiene and Tropical Medicine, Fajara, Banjul, The Gambia

Medical Affairs, Vaccines and Antivirals Pfizer, Paris, France

Jennifer Moïsi

Department of Para-Clinical Sciences, The University of the West Indies St Augustine Campus, St Augustine, Trinidad and Tobago

Michele Nurse-Lucas & Patrick E. Akpaka

Hacettepe University Faculty of Medicine, Department of Medical Microbiology, Ankara, Türkiye

Özgen Köseoglu Eser

Department of Infectious Disease Epidemiology, Faculty of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, London, UK

Anthony Scott

Centre for Genomic Pathogen Surveillance, Big Data Institute, Nuffield Department of Medicine, University of Oxford, Oxford, UK

David Aanensen

School of Public Health, Imperial College London, London, UK

Nicholas Croucher

European Molecular Biology Laboratory, European Bioinformatics Institute, Wellcome Genome Campus, Hinxton, UK

John A. Lees

Department of Biostatistics, Faculty of Medicine, University of Oslo, Oslo, Norway

Rebecca A. Gladstone & Gerry Tonkin-Hill

Department of Epidemiology of Microbial Diseases, Yale School of Public Health, New Haven, CT, USA

Chrispin Chaguza

Institute of Microbiology and Infection, College of Medical and Dental Sciences, University of Birmingham, Birmingham, UK

David Cleary

Pneumonia and Pandemic Preparedness, Global Health Division, Bill & Melinda Gates Foundation, Seattle, WA, USA

Keith P. Klugman & Gail Rodgers

Department of Pediatrics, Division of Pediatric Infectious Diseases, School of Medicine, University of Utah, Salt Lake City, UT, USA

Anne J. Blaschke & Nicole L. Pershing

You can also search for this author in PubMed   Google Scholar

The Global Pneumococcal Sequencing Consortium

• Alejandra Corso
• , Paula Gagetti
• , Abdullah W. Brooks
• , Md Hasanuzzaman
• , Samir K. Saha
• , Senjuti Saha
• , Alexander Davydov
• , Leonid Titov
• , Samanta Cristine Grassi Almeida
• , Paul Turner
• , Chunjiang Zhao
• , Margaret Ip
• , Pak Leung Ho
• , Pierra Law
• , Jeremy D. Keenan
• , Robert Cohen
• , Emmanuelle Varon
• , Eric Sampane-Donkor
• , Balaji Veeraraghavan
• , Geetha Nagaraj
• , K. L. Ravikumar
• , J. Yuvaraj
• , Varun Shamanna Noga
• , Rachel Benisty
• , Ron Dagan
• , Godfrey Bigogo
• , Jennifer Verani
• , Anmol Kiran
• , Dean B. Everett
• , Jennifer Cornick
• , Maaike Alaerts
• , Shamala Devi Sekaran
• , Stuart C. Clarke
• , Benild Moiane
• , Betuel Sigauque
• , Helio Mucavele
• , Andrew J. Pollard
• , Rama Kandasamy
• , Philip E. Carter
• , Stephen K. Obaro
• , Deborah Lehmann
• , Rebecca Ford
• , Theresa J. Ochoa
• , Anna Skoczynska
• , Ewa Sadowy
• , Waleria Hryniewicz
• , Weronika Puzia
• , Sanjay Doiphode
• , Ekaterina Egorova
• , Elena Voropaeva
• , Yulia Urban
• , Tamara Kastrin
• , Kedibone Ndlangisa
• , Linda De Gouveia
• , Mushal Ali
• , Nicole Wolter
• , Cebile Lekhuleni
• , Carmen Muñoz Almagro
• , Alba Redin Alonso
• , Desiree Henares
• , Somporn Srifuengfung
• , Brenda Kwambana-Adams
• , Ebenezer Foster-Nyarko
• , Ebrima Bojang
• , Martin Antonio
• , Peggy-Estelle Tientcheu
• , Jennifer Moïsi
• , Michele Nurse-Lucas
• , Patrick E. Akpaka
• , Özgen Köseoglu Eser
• , Anthony Scott
• , David Aanensen
• , Nicholas Croucher
• , John A. Lees
• , Rebecca A. Gladstone
• , Gerry Tonkin-Hill
• , Chrispin Chaguza
• , David Cleary
• , Kate Mellor
• , Bernard Beall
• , Keith P. Klugman
• , Gail Rodgers
• , Paulina A. Hawkins
• , Anne J. Blaschke
•  & Nicole L. Pershing

Contributions

S.B., H.S. and S.D.B. conceived the study. S.N., S.D., A.v.G., S.A.M., L.M., M.d.P. and S.D.B. collected and contributed the genomes from South Africa, and the GPS Consortium contributed other global genomes. S.B. collated the genomes and ran bioinformatics pipelines. N.L. developed the fitness framework. S.B., N.L., H.S. and S.D.B. performed the analyses. S.B., A.v.G., M.d.P., S.D., S.A.M., S.W.L., S.D.B. and H.S. discussed results and helped to contextualize them within a South African context. S.B., H.S. and S.D.B. wrote the first draft of the manuscript. All authors reviewed the first draft of the manuscript with special attention by S.B., N.L., S.N., S.D., M.d.P., S.W.L., L.M., S.A.M., A.v.G., S.D.B., H.S., A.P., C.M.A., K.K., J.L., N.C., D.L., R.K., R.D., S.M., B.V. S.B., N.L., H.S. and S.D.B., discussed the results and contributed to manuscript revisions. Funding was acquired by S.D.B., H.S. and S.B.

Corresponding author

Correspondence to Sophie Belman .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Peer review

Peer review information.

Nature thanks Taj Azarian, C. Buddy Creech, Nick Ruktanonchai and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended data fig. 1 time resolved trees for dominant gpscs..

Trees are recombination masked, aligned to a reference for each GPSC. Time resolution was performed using BactDating. The dates are along the x-axis. (a) GPSC79, N = 102 (b) GPSC68, N = 97 (c) GPSC17, N = 531 (d) GPSC14, N = 521 (e) GPSC13, N = 611 (f) GPSC10, N = 718 (g) GPSC5, N = 841 (h) GPSC2, N = 1430 (i) GPSC1, N = 1943.

Extended Data Fig. 2 Risk ratio framework sensitivity analyses to determine geographic structure when sub-sampling and only including disease isolates.

Risk ratio of being the same GPSC within a province (blue), between different provinces over increasing distance (red), compared to geographically distant pairs (>1000 km) (reference) (top) and the risk ratio of having a tMRCA 0–5, 5–10, 10–20, or 20–200 years ago within South African provinces (blue), across larger distances within South Africa (red), from South Africa to other countries in Africa (green), and from South Africa to countries outside of Africa (purple) (South Africa; N = 6910). (a) Including all isolates from each province (left) compared with sub-sampling to 300 with replacement to compensate for biased sampling in each province. (b) Including only isolates sampled from patients with pneumococcal disease. All plots use a reference of pairs which are from distant provinces in South Africa (open triangle). Error bars represent 2.5 to 97.5 CIs.

Extended Data Fig. 3 Relative risk of a pneumococcal strain being in each of municipality after 1 year of transmission.

Sequential transmission chains starting in municipalities with (a) <50 people/km2, (b) 50–500 people/km2, or (c) >500 people/km2 across 100,000 samples.

Extended Data Fig. 4 Estimated mobility and proportion of infections.

(a) a gravity model with two parameters adjusting the destination population size (beta) and the distance between locations (gamma) (pink), (b) a distance model whereby the probability of mobility is a function of the distance between locations adjusted with parameter gamma, (c) is the Meta mobility data between locations with a parameter adjusting the probability of staying in the home location (main model) across distances. (d-f) are the estimated proportion of infections by each of these models compared to the population size.

Extended Data Fig. 5 Replicated simulations for model performance testing.

Testing model performance using replicated simulations. (a) Simulated the total epidemic (black dots), biased down-sampled data as per true proportion per province (red dots) (“Biased Down-sample” in c and d), model fit to down-sampled data (red line), removing the sampling probability the model recapturing the true epidemic (black dot) (b) Population size (x-axis) compared to the proportion of infections from the down-sampled data (black) compared to the truth from the overall simulated epidemic (purple). (c) The probability of being in the home municipality and (d) the recaptured parameter after inputting a parameter of −2 to adjust the diagonal of the mobility matrix after one transmission generation. Both c and d include values from left to right for sampling as per the true data proportions in each province (6.5% of total infections), down-sampling to fit on only 2 of the 9 provinces, and if our generation time estimate is 50% smaller than the truth, exactly right, or 50% larger. Error bars represent 2.5 to 97.5 percentiles.

Extended Data Fig. 6 Comparison of fitness model results with full data or disease only data.

(a-b) Results with the full data. (c-d) Results with the disease-only data. a and c present the model fits for the proportion of serotypes from non-vaccine type (NVT), PCV7 types, and additional PCV13 types not included in PCV7 from the years 2000 to 2014 in this study. Points represent data and line represent the model fit. b and d present the relative fitness estimates for all three groups of serotypes in each era. Pre-vaccine era is prior to 2009 for NVTs, prior to 2009 for PCV7 and prior to 2011 for PCV13. Post-vaccine era is post-2009 for NVTs, post-2009 for PCV7 and post-2011 for PCV13. Error bars represent 2.5 and 97.5 percentiles.

Extended Data Fig. 7 Fitness growth model testing year of switch and schematic.

(a) Testing year of fitness switch for the logistic growth -fitness model. Adjusting the year of the fitness switch in the model fitting to vaccine status. The difference to the best WAIC (2009 [PCV7 implementation] & 2011 [PCV13 implementation]) is on the y-axis where the year of fitness switch relative to 2009 & 2011 is on the x-axis. Further we test no fitness switch (ns; yellow) and the impact of including one fitness switch in 2009 (purple). The dark gray box highlights equivalent models (ΔWAIC ≤2) and light gray box highlights similar models (ΔWAIC ≤ 7). (b) Schematic denoting the fitness growth model parameterisation which accounts for the specific timing of the PCV impacting each group of serotypes (NVT in blue; PCV7 in green; PCV13 in red).

Extended Data Fig. 8 Serotype fitness estimates.

Fitness estimates pre- and post-PCV (y-axis) for each serotype (grey), superimposed by group including NVT (blue), PCV7 (green), and PCV13 (red). Pre-vaccine and post-vaccine refer to pre- and post- 2009 and 2011 for PCV7 and PCV13 respectively. Individual serotype fitness estimates can be found in Fig. S 9 .

Extended Data Fig. 9 Antimicrobial resistance summary.

The proportional trends in antimicrobial resistance overall (black) and within Vaccine type [VT] (in blue) and Non-Vaccine Type [NVT] (in red) serotypes for in-silico predicted (a) penicillin (b) erythromycin (c) co-trimoxazole and (d) clindamycin.

Extended Data Fig. 10 Data fits for model accounting for proportions and fitness over time in four groups.

(a) NVT-penicillin resistant (red), (b) NVT-penicillin susceptible (green), (c) VT-penicillin resistant (yellow) and (d) VT-penicillin susceptible (blue). The dashed lines indicate the year of PCV7 implementation and fitness switch model implemented. (e) Fitness estimates pre-PCV and post-PCV for each group and colored accordingly. e is on a log scale. This model uses a shift in fitness in 2009. Error bars represent 2.5 to 97.5 percentiles.

Extended Data Fig. 11 Data descriptions across the 234 municipalities of South Africa.

(a) Population density as estimated given the area of each municipality and the populations estimated by LandScan and (b) the radius of gyration for each municipality given the distance between each municipality at the centroid weighted by the human mobility data from Meta Data for Good.

Extended Data Fig. 12 Parameter adjustment sensitivity analysis.

(a) Fitting the mobility model to 15 years of evolutionary distance. Fitting the probabilistic mobility model to pairs of genomes which are 15 years divergent from their MRCA (black), compared against the 10 years used in the main model (red), and the data (blue). (b) Assessing the mean geographic distance per evolutionary time in years for the data (blue), including the sampling probability (red) for, (left) generation time of 15 days, (middle) generation time of 35 days, and (right) generation time of 55 days.

Supplementary information

Supplementary information.

This file contains Supplementary Figs. 1–20 and Supplementary Tables 1–13.

Reporting Summary

Peer review file, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Belman, S., Lefrancq, N., Nzenze, S. et al. Geographical migration and fitness dynamics of Streptococcus pneumoniae . Nature (2024). https://doi.org/10.1038/s41586-024-07626-3

Download citation

Received : 31 May 2023

Accepted : 30 May 2024

Published : 03 July 2024

DOI : https://doi.org/10.1038/s41586-024-07626-3

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

TOP > Projects > Data-based Discussion on Education and Children in Japan > Data-based Discussion on Education and Children in Japan 6: To prevent "cycle of poverty" - current situation in Japan concerning continuation of education to college

• Data-based Discussion on Education and Children in Japan 6: To prevent "cycle of poverty" - current situation in Japan concerning continuation of education to college

Data-based Discussion on Education and Children in Japan

• 1. Bukatsudou --A Puzzling Activity that Most Students Join
• 2. Analyzing Juku --Another School After School
• 3. Low Self-Esteem Among Japanese Children--How to Overcome This Issue
• 4. Issues Regarding Japanese Children's Reading--In Association with the Effects of Reading
• 5. Now's the time amid COVID-19 fear, let's close up on the positive side of a "well-regulated lifestyle"
• 6. To prevent the "cycle of poverty" -current situation in Japan concerning continuation of education to college (This paper)

The issue of "Poverty" in Japan

In 2014, the Japanese people were confronted with the shocking news that one in six children were living in poverty. In that year, the Ministry of Health, Labour and Welfare announced that 16.3% of children under the age of 18 lived in households whose income ranked below half of the median household income, which was the worst record in history (Figure 1). This figure later improved to 13.9% according to research conducted in 2015, but it still remains at a high level, putting Japan among the high poverty rate nations of OECD members.

Japan has been called as a society of "all Japanese belong to the middle class." The dominant sentiment is that most Japanese people live above a certain standard of living, even though there are few very rich people. In the "Public Opinion Survey on the Life of the People" conducted by the Cabinet Office, 90% of the respondents rated their living standard as "middle class." This trend has remained unchanged from the 1960s until the present. In our daily lives, we do not see people who are very poor, without housing or food. Therefore, we are stunned at the revelation that the number of children living in poverty is increasing without our knowledge.

However, the signs were already making an appearance, and schoolteachers became aware that families in financial difficulty were increasing in number. That was shown by the increase in the number of children receiving school expense subsidies. Since the latter half of the 1990s, there has been a large increase in the number of children eligible for school expense subsidies. Before 2000, the proportion of children who qualified for and received school expense subsidies due to low family income was less than 8%, while it almost doubled to 15.6% in 2012 (Figure 2). This almost matches the result of the Comprehensive Survey of Living Conditions introduced at the beginning. Theoretically, around six children in a class of 40 are applicable.

Researchers have also revealed that the "education gap" comes about due to economic conditions and the cultural background of families. Traditionally it was considered in Japan that the gap between the rich and the poor would diminish due to equal distribution of good quality education. Many Japanese shared the belief that even a poor child could become successful in society if he/she performed well in schools by applying his/her ability and effort.

However, research in the 2000s eradicated such belief. In educational sociology research, my expertise, it has been revealed that gaps exist in the performance of children at school or their later education in colleges. Depending on the economic circumstances and culture of their families, a class difference exists in their motivation toward social success. Moreover, it has been proved that this gap has been widening in recent years *1 . What is indicated is a larger possibility that the effect of school education is to reproduce and enlarge the gap between rich and poor (gap enlarging function) rather than to have a shrinking effect (equalizing function). It is comparatively easy for children of affluent families to maintain or improve their position through education, while poor families are unable to give sufficient education to their children. As a result, the gap tends to widen.

The government has not been inactive in this situation. In 2013, the "Law on Measures to Counter Child Poverty" was unanimously passed in both Houses of the Diet. In 2014, the "General Rule on Measures to Child Poverty" was approved by Cabinet decision. In the General Rule, it is clearly written that the indices concerning child poverty are established in specific terms, and the ministries and agencies in charge cooperate to comprehensively promote the measures. Various viewpoints to correct the gap were included such as guaranteeing academic ability, reducing the burden of educational expenses, support for living and recruitment of guardians, and financial assistance. Furthermore, the "Law to Support Education in Colleges and Other Institutions" was passed in May 2019, providing low income families with exemption/reduction of tuition and offering scholarships.

Gap in continuing education into college

Such efforts should be continued, however, they are insufficient to solve the current problems, one of which is the gap in continuing education into college.

In Japan, the high cost of higher education discourages children in poverty from going on to receive college education. Figure 3 shows the ratio of higher education bearers in proportion to GDP in each country. This indicates Japan is positioned in the middle among OECD countries for expenditure on higher education. However, a large amount is paid privately while public expenditure stays low. Public expenditure in proportion to GDP is at the lowest level. The government simply does not spend sufficient money.

This means that the burden weighs heavily on families. It is not rare to see the expense of higher education exceeding 10 million yen when the cumulative expense concerning preparation for college is added, plus housing and living expenses away from family, etc., on top of enrollment fees and college tuition. If there is more than one child in a family, the burden weighs even heavier. When asked the reason why couples do not realize their ideal number of children, many couples answered "It costs too much to raise and educate children" ("Japanese National Fertility Survey" by National Institute of Population and Social Security Research). It is widely recognized that education costs too much.

Let us see the difference in the ratio of continuing education in colleges by household income. Figure 4 shows the result of a survey asking third-year senior high school students what they would do after graduation, divided by household income. In Japan, somewhat fewer than 60% of students go on to colleges or two-year colleges, so caution is necessary in that the population of this research includes more than the average response for higher education. Still, it is obvious that household income affects the difference in the path choices after graduation. The lower the household income, the more students choose "vocational school" or "employment," while the higher the income, the more students choose "college" or "preparation for college." The total of responses for "college" and academic "preparation for college" exceeds 90% among households whose income is "10 million yen and above," while it is lower than 50% in households whose income is "less than 3 million yen." Children of low-income families have relatively fewer opportunities to receive higher education. It should be added that children of higher income families tend to go to topnotch schools among those who go on to colleges. For example, "Survey of Living condition of Students" conducted by The University of Tokyo clearly shows a deviation in the household income of students' families toward higher class. The tuition fees of national universities should be affordable, but that does not mean students from low-income families are enrolled.

Problem of family environment in the background

The graph on the top in Figure 5 shows the response to the statement "I want my child to get better grades so that he/she is admitted to the best university possible." You can see that the lower the annual income of the parents, the more negative the response, while positive responses increase as income becomes higher. Similarly, the bottom graphs show the responses to the statement "I want to spend money on education for my child even if we have to overstrain ourselves a little." Again, the households with higher income give more positive responses.

As you can see, household income and awareness of education are correlated, which is reflected in actual behavior. Seen relatively, the low-income class contains more parents who are not enthusiastic about education. They tend to show passive behavior in educational encouragement to their children or educational investment outside schools. As a result, their children are less able to be motivated for continuing education in colleges. They avoid study, let their grades drop, and make choices other than going on to colleges.

Should be viewed as a structural problem

When considering the relationship between the socio-economic status of parents and choice of career path by children, "Learning to Labour: How Working Class Kids Get Working Class Jobs" by Paul E. Willis is a good source of reference. This research was conducted in the UK in the 1970s under the important theme of class reproduction through school education.

This pioneering research reveals that there is a paradoxical structure of existing social system reproduction by children of the labourer class denying school education (meaning middle-class culture). It should be noted that being one of the "lads" involves actively choosing to fall off from the system and not receive college education. Their choice lands them a job in factories (workplace resembling prison cells without freedom) and ultimately absorbed into the social system. Only "education" can set them free from the prison, but it is too late by the time they realize that.

In Japan, class division is not that obvious and cultural clash is not apparent. This point shows a palpable difference from the situation in the UK depicted by Willis. Also, opportunities for college education are given equally to high school graduates (or those with equivalent education). It appears to be equality of opportunity. This presents all the more reasons for the children themselves and people around them to conclude it was the results of their own choices that children in poverty did not go on to study in colleges. However, it may be due to the mechanism that leads them "not to choose colleges" - the mechanism that prevents the introduction of the value of higher education and cools their motivation for studying.

We should not only blame the children for being responsible for their choice but also look at this problem as a structural one, where family environment and school education are making an impact in complex ways. We also need extra consideration for children in poverty to prevent school education from enhancing the problem. It also brings huge merit to society to teach them that education is one of the ways to get out of the prison called poverty.

Breaking the cycle of poverty

Japan has a problem of productivity too slow to be improved in addition to its rapidly aging society. We cannot maintain our national strength in this way. Some developed countries suffer from similar problems. It is considered important to continuously support childbirth and child rearing (and work-style support along with it) as well as enrichment of education. The cycle of poverty works negatively on both of those.

The current situation of an increasing poverty rate among children was introduced at the beginning of this paper. If unattended, a further declining birthrate and social inactivity will result. First, it will be difficult for people in poverty to get married, have children and raise them. Also, in a situation where the gap is reproduced through education, the class who feel they are not receiving the merit of education will be reluctant to actively receive education. It is thought to be the cause of lower productivity throughout the whole of society and result in higher cost of social security, etc.

Breaking the cycle of poverty will lead to the creation of a society where success gained by ability and effort is justly rewarded, not success bestowed by birth. This is an ideal principle in modern society. Are we getting near to the ideal? I think it is about time we, not only in Japan but also all countries and regions, everyone, should take it seriously and deal with this problem beyond our borders.

• *1 Takehiko Kariya, (2011). Kaisouka nippon to kyouikukiki -fubyoudousaiseisan kara iyoku kakusa shakai (insentibu dibaido) [Class divided Japan and Educational Crisis - from Reproducing Inequality to Motivation Gap Society (Incentive Divide)] Yushindo.

• Thinking about children's issues from the perspective of poverty
• Revisiting the Gender Gap: Japanese Mothers' Views of Their Children's Education, Work, and Future
• Child Poverty and Child-Caring Design in Social Welfare
• Data-based Discussion on Education and Children in Japan 5: Now's the time amid COVID-19 fear, let's close up on the positive side of a "well-regulated lifestyle"
• Data-based Discussion on Education and Children in Japan 4: Issues Regarding Japanese Children's Reading--In Association with the Effects of Reading

Latest Posts

• [Aiming to Realize an Inclusive Society] An Inclusive World Realized by Art II: What Can We Do to Make Their Day in the Way They Desire?
• [Aiming to Realize an Inclusive Society] An Inclusive World Realized by Art I: We Do Not Want to Let Them Be Someone They Do Not Want to Be
• [Snuggling Up to Our Differences] Episode 8: There Is No Such Thing as a Bad Child
• [Parenting and Childcare in Germany - Insights from Berlin] Inclusive Education in Germany: A Japanese Perspective on Practices, Challenges, and Pathways to the Future
• ISS BERD Joint Research Project (JLSCP) Results of "2023 Fact-finding Survey on Children's Utilization of ICT": About 80% of School Children Enjoy Classes Using ICT Devices [Part2]
• ISS BERD Joint Research Project (JLSCP) Results of "2023 Fact-finding Survey on Children's Utilization of ICT": About 80% of School Children Enjoy Classes Using ICT Devices [Part 1]
• [Snuggling Up to Our Differences] Episode 7: Avoid Burying the Fun of Learning in Competition
• [Aiming to Realize an Inclusive Society] Providing an Environment Where Children Can Learn Things Independently and Proactively, Seeking to Enhance the "Ability to Become Happy"
• [Japan] Identifying Factor Structures and Determinants Involved in the Development of "Non" Cognitive Skills in Early Childhood
• Seven-Year Follow-up Survey Data on the Reading Behaviors of Children from Elementary to High School: About 50% of children spend 0 min of their day reading books [Part 2]
• Early Childhood
• Elementary, Junior High, High School, and University
• International Comparisons
• Digital Media and Children
• Children's Rights and Well-being
• New Directions
• Full Paper Archive
• Research Data
• Latest Children's Issues
• Childrearing and Education
• About this Project
• Interactive Activities
• CRN History
• Publications

Mobile Navigation

Find it fast container, horizontal navigation, in this section, education : poverty in japan 4 of 4.

*This "Poverty in Japan" series of articles is based on a presentation given by the Seisen High School Social Justice Committee to their peers at an assembly in November 2021.

1) The Prevalence of EDUCATION INEQUALITY in Japan

Basic stats & myth busters.

Why is educational inequality prevalent in Japan, and how does poverty contribute? We must begin with an understanding of what constitutes inequality in education.

What is Education Inequality?

Educational inequality is the unequal distribution of academic resources. This can include a variety of aspects to academic life from extracurriculars to tutoring outside of school, and even cram school. A primary example of this throughout Japan these past two years was the impact of COVID on impoverished families lacking in technology and means of connection while we were able to continue classes online through zoom.

On this next slide, we will address common misconceptions and perceived notions that might be masking the impact of poverty on education within japan. While we may believe that a relative majority of the population can read and write, the 99% literacy rate idea is false. Though it may have been accurate in 1948, such is no longer applicable. Though there is no up-to-date government data, there are signs that illiteracy may be more common than Japanese people believe. The most recent census, conducted nine years ago, found that more than 128,000 people had left school by the age of 12, mainly due to a lack of financial resources.

And circling back to education during COVID, Educational disparities and inequalities have been felt more and more by younger generations.With survey results showing More than half of young adults in Japan felt there were disparities between students in access to learning opportunities amid school closures , carried out by the Tokyo-based nonprofit organization Nippon Foundation, finding 58.6 percent of respondents age 17 to 19 felt there were inequalities in education.

So, while Every citizen in Japan is guaranteed to obtain the right to education. However, interviews organized by NHK proved the opposite. From their responses, some have missed elementary or junior high; unable to solve simple math or write hiragana. Unable to follow basic instructions, or read announcement signs in public due to illiteracy.  A true story and example of a student our age is Kosuke - a 19 year old living in eastern Japan.  Due to abuse and lack of financial support from a single mother who had fallen ill, he had entered into the working field able to read but unable to write aside from his own name and address. Leading to difficulty completing tasks within the workplace, many face similar situations throughout Japan today

Lemurik. "Child Abuse Domestic Violence Vector Illustration Stock Vector (Royalty Free) 1321160498." Shutterstock.com. n.d. Web. 17 Nov. 2021. <https://www.shutterstock.com/image-vector/child-abuse-domestic-violence-vector-illustration-1321160498> Marina, Shirakawa. "Educational Poverty in Japan | NHK WORLD-JAPAN News." NHK WORLD. 6 Apr. 2020. Web. 17 Nov. 2021. < https://www3.nhk.or.jp/nhkworld/en/news/backstories/226/ >  

2. What education Inequality can we see around us?

So now that we acknowledge the existence of education inequality in Japan, what type of education inequality can we see around us?

Escalator Schools

Have you ever heard of an escalator school system in Japan? As you might be able to easily catch it from the name itself, it is a school that automatically offers the student to move on to the next education level, all the way up to the university, IF the student is accepted for the same institution’s elementary school, or sometimes kindergarten. Although the ratio differs for every school, the parent university may automatically accept as many as the top 50 or 60 percent of the students from the same institution’s high school. Keio University, one of the Japan’s top university, accepts over 95 percent of their own high schools’ graduates. Instead, the entrance test for the elementary school or kindergarten is notoriously difficult. Other highly selective private academic kindergartens within the escalator system can be Sacred heart, Keio, Waseda, Gakushuin, and Aoyama Gakuin. 

This escalator school is an example that can create education inequality related to poverty as children from poor families are difficult to take the test. The children are half expected/forced to go to jukus to prepare for the kindergarten entrance exam and most likely, the parents of the poor families wouldn't be able to afford such money to prepare for the test and also for the tuition of a private school that will continue for around 16 years or more. Moreover, the private academic kindergartens I listed earlier not only tests various skill and intelligence tests to the children applying for admission but also to their parents!  

Urgency of Starting Young

The majority of Japanese students start their education from as young as three years old. Their parents would fight to enroll them into a school they saw fit. Extracurricular classes such as piano or swimming lessons became popular amongst these families as those two were also being taken by Tokyo university students. Some families also valued learning English and would submit their children to English speaking preschools.

Meanwhile, the richer families who place a lot of importance on education face a challenging examination starting in elementary school as soon as the student turns the age of six. Despite being so young, they would have had years of studying and preparation for this. Instead of looking at public schools, these parents seek to send their children to the elite private schools that focus on a higher standard of education. These entry exams are more than just a simple interview and test, instead they are tested for things such as seasonal knowledge with written tests, being put in a group of children to see how well they behave with other students, and involving them in handicraft to evaluate their comprehensive abilities and hand coordination. The tuition for schools like this are very heavy as it could reach 1,863,085 yen minus the costs of other supplies as well as donations made throughout the year.  This itself creates educational inequality as families who may not be able to afford such tuition would be overlooked and missing out on different opportunities. If families with a lower financial income were looking into enrolling their children into prestigious preschools, they would need at least a couple of years of studying beforehand to be fully prepared for the examination tests. Kumon being the most prominent correspondence course students would be taking, their families would then have to pay for those lessons as well as the extracurricular activities they would be taking outside of school. With all that, the price steadily increases and is hard to keep up with. This itself creates an unfair division between the families and more specifically the opportunities handed to the students as the acceptance into an elite school starting from the age of six is what sets the educational trail for them. If accepted into a “lower” public school, it would affect the next school they would attend and the chances given to them as they reach university. 

As we will later on explain, these opportunities and acceptances to the private schools are affiliated with some elite colleges within Japan and is why there is a lot of competition to attend these kinds of schools.   

3. Cycles of poverty & Education inequality

Cultural impact on education poverty.

Before we get into the actual cycle, I want to get into the specific reasoning on why this cycle happens in the first place.  Japan values the ‘label’ of the school that you go to. Hence if you go to a school with good representation, there is a higher chance you will be able to get into a better company or job position. According to FRaU, even simply going to college or not makes a 40% difference in pay. (FRaU) This creates the hierarchy between people that do and do not have the financial support to go to a good school, leading to a poverty cycle for the ones who don’t have that support.   

The Cycle of Poverty + Education

Here is the cycle: It goes, a child from a wealthy family is born → the child can go to better facilitated schools, hence receiving better education → ending with the child can also getting better jobs and pay Meaning, that when you don’t have good pay, there is a low possibility of getting into good schools after middle school and to receive good pay. Even when considering the option of scholarships, most scholarships are low price and provided only for universities, and the amount of schools that provide this is only 50%. Even ones that are provided by the government usually don’t provide for higher end schools. (奨学金.net) Even with large amounts of effort, there is the reality in which you can’t develop your skills because of a financial barrier. 

A Minority Group Poverty Cycle...

Another cycle is one between minority groups. There is a lack of support for minority groups → which leads to a lack of education → ultimately leading to poverty There are still a lot of stereotypes that come with minority groups like the disabled, so most public schools can’t be completely cooperative because of the ‘extra fuss.’ (朝日新聞デジタル 2019) This leads to lack of education in the minority, providing even less opportunities than the average abled person, and has a higher rate of poverty. (leading back to cycle) The job acceptance rate for the disabled in 2021 was 42.2%, very low compared to the 96% acceptance rate of the abled. It is also researched that the poverty rate of the disabled goes a little over 25% of the whole disabled population, which means 1 in 4 disabled person is in poverty. (日本経済新聞; イノベーションズアイ) This problem’s solution will be addressed in the next section, but in general as students, it is important to know that there is a clear cycle of education and finance that is ultimately leading people to poverty in Japan. 

4. What action can we take as individuals and a community?

Organizations.

https://kidsdoor.net/

KIDSDOOR is a non-profit organization engaged in various support activities for children in Japan

• The group aims to stop the poverty cycle,
• Where children grow up to become poor because their parents’ poverty kept them from accessing a sufficient education 
• They have been conducting a study support program “dubbed Gakubora” since August 2009, and it is intended to give educational assistance to children who go to public school or are part of a single-parent family, needy family, or orphanage by widely mobilizing volunteers to help eliminate unequal educational opportunities for children across the nation. 
• College students are enlisted as volunteer tutors to extend help to children who don’t have access to education due to their poor static

https://learningforall.or.jp/

Learning for All  is an organization supporting children in Japan who struggle with poverty, which ultimately affects their education.  Money that is donated to this organization is used for transportation fees for student volunteer tutors, sketchbooks, files, textbooks, and venue expenses for public halls.

Real Life Example

Raising awareness on the topic, like we are now about education inequality due to poverty is one of the best things we can do. Citizens of Japan don’t have much control over the issue so more people joining organizations and signing petitions can make a difference.

“At my previous school the Student Council raised money to donate to an organization for the homeless community in Yoyogi park. The money we raised went towards providing Christmas eve meals and other supplies such as emergency blankets, and non perishable food.”

Mental Health: Limited resources to diagnose mental health issues at a young age lead more people into poverty each year. One of the main organizers of Grama Seva(the support group for the homeless in Yoyogi park) said that “Most people we see coming in for supplies show signs of mental illness, I think if mental health was more of an open conversation in Japan, fewer people would have ended up here." Japan's societal expectations need to shift for people with mental health issues to be accepted, undergo treatment and find their place in society, so that these people do not fall into the poverty cycle. 

Works Cited

Lemurik. "Child Abuse Domestic Violence Vector Illustration Stock Vector (Royalty Free) 1321160498."  Shutterstock.com. n.d. Web. 17 Nov. 2021.  <https://www.shutterstock.com/image-vector/child-abuse-domestic-violence-vector-illustration-1321160498>

LIVE JAPAN. "Behind Japan’s Elite Education—Insight Into its System | LIVE JAPAN travel guide." LIVE JAPAN. n.d. Web. 18 Nov. 2021. < https://livejapan.com/en/in-tokyo/in-pref-tokyo/in-tokyo_train_station/article-a0000955/ >

Marina, Shirakawa. "Educational Poverty in Japan | NHK WORLD-JAPAN News." NHK WORLD. 6 Apr.  2020. Web. 17 Nov. 2021. < https://www3.nhk.or.jp/nhkworld/en/news/backstories/226/ >

Masakazu Hojo. "Inequality in Japanese Education." Taylor & Francis. n.d. Web. 17 Nov. 2021.  < https://www.tandfonline.com/doi/abs/10.2753/JES1097-203X360301 >

The Japan Times. "Over half of young adults in Japan felt education gap as schools closed over  coronavirus." The Japan Times. 19 Jun. 2020. Web. 17 Nov. 2021.  < https://www.japantimes.co.jp/news/2020/06/19/national/education-gap-schools/ > Wray, Harry. Japanese and American Education: Attitudes and Practices. Bergin & Garvey, 1999.    

Seisen Social Justice Committee "Poverty in Japan" Article Series

• Introduction
• Food Sustainability
• Fashion Sustainability
• Immigration

More Seisen News

Michi Morita (Class of 2025): "Along with these titles, I also received scholarships to summer training programs in New York at Steps on Broadway. These titles and scholarships not only gave me amazing opportunities to network with professionals, but it also gave me some amazing friendships that I will have forever."

Mihika Gupta (Class of 2029) "As I compete for our school during the various sports seasons, I seize the opportunities to catch up with my equestrian training during the summer or in between sports seasons."

Rena Kenmoku (Classof 2024): "Music should be something that you laugh, bleed, and breathe out. So when you compose it should sound like a reflection of yourself, be original, and don't let anybody influence you to do otherwise."

A group of High School students conducted sustainable tourism research in Thailand.

Sophia Migdalski (Class of 2025): " When you're having fun and being genuine on stage, whether you're singing or engaging in any other activity, the audience will naturally have a good time watching you."

Natsuki Uchino (Class of 2024) "Within sports, mental strength—not only physical—is always something you need to work on. There will always be ups and downs in your performances, and it is important to know that it is normal"

Map Options

Poverty Rate by Country 2024

Poverty can be cyclical, with lack of access to education, healthcare, and job opportunities perpetuating the inability to improve financial situations.

South Sudan has the highest poverty rate at 82.3%, indicating severe economic challenges and a need for significant humanitarian and developmental aid.

The United States, with a significant wealth inequality gap, has varying poverty rates, up to 17.8%, pointing to challenges even in the world's largest economy.

Poverty is a state of being in which a person lacks the income (or other means of support) to reliably meet their basic personal needs, such as food, shelter, and clothing. Poverty exists in every country in the world, though it is a more pressing issue in some countries than in others. The poverty rate is the number of people (usually expressed as a percentage) in a given demographic group whose income falls below the poverty line.

Poverty has a wide range of possible causes, from the amount of fresh water and arable land in a region to government policies or ongoing armed conflict. Additionally, natural disasters such as the COVID pandemic or the 2020 earthquakes in Puerto Rico can further strain an impoverished area's already scarce resources.

Poverty can be a cyclical trap. For people to rise above poverty, they need education , proper health care and sanitation, access to clean water, and job opportunities that can help them improve their financial situation. Unfortunately, people in poverty often live in areas low on these resources. Therefore, the people become trapped in a vicious cycle in which they can't get better jobs until they improve their situation (education is particularly helpful), but they can't afford to improve their situation until they get better jobs.

To help break the cycle of poverty, organizations including the United Nations , World Vision , and Global Citizen have worked alongside various governments to improve impoverished people's access to clean water, adequate food, affordable education and health care, and other needs.

Country-wide poverty is typically measured in one of two ways. The first is to determine the percentage of people whose daily income falls below specific baseline amounts, such as $10 per day. These baselines remain the same for every country, enabling a slightly different perspective on country-to-country comparisons.

The most widely used baseline amount is $1.90 per day, measured in 2011 PPP international dollars (INT), a theoretical unit of currency used to make country-to-country comparisons easier. People who make less than this amount are considered to be in "extreme poverty" , which is to say they are the poorest of the poor. Additional baselines such as $3.30/hr and $5.90/hr are often employed to help count people whose poverty is slightly less extreme.

The second way to measure a country's poverty level is to determine the percentage of people or families who earn less than the "national poverty line", or poverty threshold —meaning, the annual income below which a person or family is considered impoverished. The national poverty line is calculated independently for each country because each country's economy is different. For example, a person earning $25,000 a year in the United States would have different opportunities than a person who earned $25,000 a year in Somalia .

10 Countries With the Highest Poverty Rate

The majority of countries in the world, as well as organizations such as World Bank , the OECD (Organization for Economic Co-operation and Development), and the European Union , set the national poverty line at 50% of a given year's median income. For instance, the median income in the United States was $67,521 in 2020 so the national poverty line according to the United Nations would be $33,761.

However, some countries use different calculations. The U.S. itself, for example, employs a formula first devised in the 1960s, which calculated the estimated cost of adequate food for a year and multiplies by three to account for additional costs (housing, utilities, medical expenses, etc). For 2020, this formula set the poverty threshold at $26,246 for a family of four —significantly lower than the more widely used World Bank method.

This discrepancy—coupled with the fact that those above the poverty threshold are often ineligible for aid programs—is the main reason that many anti-poverty advocates and policy experts including the Washington Post argue that the U.S. formula is badly outdated and has failed to keep up with the rising cost of housing and other expenses.

Globally speaking, the number of people living in extreme poverty has been on the decline for several decades, from 1.94 billion in 1982 to 696 million in 2017. This decrease is particularly encouraging because the Earth's population rose considerably during this same time period, from roughly 4.5 billion people in 1981 to more than 7.8 billion in 2021.

While overall poverty rates have improved considerably in recent decades, several individual countries have experienced a rise in poverty. As previously mentioned, 696 million people still live in extreme poverty, surviving on less than $1.90 (INT) per day. More than 430 million of these people live in Sub-Saharan Africa , the poorest region in the world, where more than 40% of people lived in extreme poverty as of 2018. Many countries in which poverty is rising have been plagued by political instability or conflict. Others are hampered by frequent natural disasters or ongoing environmental stresses (increased drought in particular) caused by climate change. Several countries in Sub-Saharan Africa face both of these concerns.

The poverty rate in the United States varies depending upon the method of measurement. According to the U.S. Census Bureau, the official 2017 poverty rate in the U.S. was 12.3% . However, other sources placed it as high as 17.8% . Despite being the largest economy in the world, the U.S. also has a significant wealth inequality gap. The 2021 poverty threshold in the United States is $26,246 for a family of four. This means that households with two adults, two children, and a pre-tax income of less than $26,246 are considered to be living in poverty. Some states are more impoverished than others, and their poverty is exacerbated by high unemployment rates and a lack of high-paying jobs.

• Poverty Rate is measured as the percentage of the population living below the national poverty line(s). National estimates are based on population-weighted subgroup estimates from household surveys.
• The World Bank updated the global poverty lines in September 2022. The decision, announced in May, followed the release in 2020 of new purchasing power parities (PPPs)—the main data used to convert different currencies into a common, comparable unit and account for price differences across countries. The new extreme poverty line became $2.15 per person per day, replacing the previous value of $1.90, which was based on 2017 PPPs.
• Poverty rates for Australia , Canada , Israel , and the United States were computed from OECD data for 2022 or the latest available year. All other countries' rates courtesy of World Bank.

Download Table Data

Enter your email below, and you'll receive this table's data in your inbox momentarily.

What country is #1 in poverty?

Frequently asked questions.

• Poverty rate - OECD
• Fact Sheet: An Adjustment to Global Poverty Lines - World Bank
• Poverty headcount ratio at national poverty lines - World Bank
• Online Analysis tool for global poverty monitoring - World Bank PovcalNet
• Poverty headcount ratio at $2.15 a day (2017 PPP) (% of population) - World Bank
• Poverty gap at $3.65 a day (2017 PPP) (%) - World Bank
• Poverty headcount ratio at $6.85 a day (2017 PPP) (% of population) - World Bank
• Share in poverty relative to different poverty thresholds - Our World in Data
• Poverty Thresholds 2020 - Census.gov
• Poverty headcount ratio at national poverty lines (% of population) - World Bank

United States Institute of Peace

Home ▶ Publications

Kenya’s Crisis Shows the Urgency of African Poverty, Corruption, Debt

KEY TAKEAWAYS

• Kenya’s protests and violence arise from unemployment, poverty and inequality.
• Supporting Kenya as a vital partner for stability is important to U.S. interests.
• Kenya reflects Africa-wide crises: deep corruption and unsustainable foreign debt.

Thursday, June 27, 2024

/ READ TIME: 6 minutes

By: Ambassador Johnnie Carson

Kenya’s public protests and deadly violence over proposed tax increases this week highlight some of the country’s most serious challenges: high youth unemployment, deepening poverty and the glaring gap between living conditions for the country’s elite and its urban poor. This social crisis is exacerbated by severe corruption, a stifling foreign debt and a too-violent response by Kenyan police, who have a poor record in handling large demonstrations. Steps to calm this crisis are vital to preserve Kenya’s overall stability, its role as an East African trade hub — and its capacity to serve as a leader for peace, which the United States increasing has relied upon in Africa and elsewhere.

This week’s eruption of protests, and the deaths or injuries of dozens of people amid clashes with police, represent the toughest crisis in the 21 months of President William Ruto’s administration. But Kenya remains East Africa’s most important and influential nation; the United States and other partners should encourage all Kenyans, in government and civil society, to resolve this crisis and help Kenya to address the deeper problems at its roots.

Protests broke out nationwide this week — in 35 of Kenya’s 47 counties, according to the Kenyan newspaper The Nation — over a budget bill in the parliament that would have increased taxes on people’s daily staples, including a 16% duty on bread and 25% on cooking oil. The protest campaign was mainly leaderless and spread largely via Kenyans’ angry discussions over social media.

Young, unemployed Kenyans were prominent in all the demonstrations. Estimates of the nation’s unemployment rate for youth vary, but the problem is undeniably acute. Thousands of school leavers and university graduates struggle every year to find employment, and young Kenyans have grown weary of government corruption and unfulfilled promises to create more jobs.

Kenya’s Critical Roles

Restoring immediate peace and sustaining Kenya’s general stability is vital first of all to the 56 million Kenyans, but is important more widely because of Kenya’s regional roles, both economically and in supporting political stability and security. Kenya is the economic and transportation hub for East Africa and parts of Central Africa. The port of Mombasa, the region’s largest, handles cargo for countries as distant as Uganda, Rwanda, Burundi and South Sudan, and eastern parts of the Democratic Republic of the Congo (DRC). Nairobi is the region’s largest financial, commercial and telecommunications center.

Diplomatically, Kenyan officials have been active in trying to resolve conflicts in Ethiopia, Sudan and eastern Congo, and its military has deployed in peacekeeping roles in the eastern DRC and Somalia. Kenya has been a close security partner of the United States and has worked closely with international efforts to counter Al Shabab and other extremist groups in Somalia and the region. Kenya has lent its airports and Mombasa’s port to facilitate U.S. humanitarian operations supplying food and medicine amid crises across East Africa: in Somalia, Sudan, South Sudan, Rwanda and the eastern DRC. Kenya hosts a small U.S. military facility in its north and America’s largest embassy in Africa.

During President Ruto’s state visit to Washington this month, President Biden recognized Kenya’s valuable cooperation on security issues by announcing that the United States had designated Kenya a major non-NATO ally. President Biden applauded Kenya’s decision to send a contingent of police officers to Haiti to help restore peace in that country — where 400 Kenyan officers arrived this week to begin that difficult mission.

Kenya’s Challenges: Debt and Corruption

Any effective, enduring response to Kenya’s turmoil must address the underlying social crises of poverty and inequality. Kenya’s economy grew well before the COVID pandemic in 2020, in part due to reforms friendly to business expansion. For years, the World Bank and others have noted, Kenya lifted many of its people out of poverty. Nearly 47 percent of Kenyans lived below the national poverty line in 2006, but Kenya reduced that rate to about 33 percent in 2019. And the current government has sustained significant growth, better than 5% last year, in part through the embrace and rapid expansion of its digital technology.

But, as in much of Africa, where improving life expectancies are driving the world’s fastest population growth, Kenya needs extraordinary and sustained economic growth to achieve its vital task of continuing to reduce poverty. A headline from the World Bank just weeks ago tells the difficult story: “ African Economies Projected to Grow by 3.4 % in 2024, But Faster and More Equitable Growth Needed to Reduce Poverty .” Any country would face a challenge in meeting the needed growth rates, but, as that report notes, many African states, like Kenya, face special burdens: the depth of past and current economic inequalities and an enormous burden of foreign debt.

Kenya also finds its growth limited by widespread corruption and some questionable economic decisions of past years. Over the past decade, Kenya has invested heavily in major infrastructure projects — a new, standard-gauge railroad from Mombasa to Nairobi, a modern international air terminal in Nairobi, major roads in and around the capital and dam projects. But these projects required heavy foreign loans. Kenya’s domestic and foreign debt now reach $80 billion, approximately 70% of its entire gross domestic product. Debt repayments now eat nearly half of the government’s budget, crimping the country’s ability to sustain the necessary development projects for growth. The debt burden pushes the country toward the instability we have seen this week — and toward the precipice of a major debt default.

For Kenya, like dozens of developing countries, debt burdens are multiplied by the global high interest rates of recent years.

For Kenya, like dozens of developing countries, debt burdens are multiplied by the global high interest rates of recent years. “The global financial architecture is no longer capable of meeting the needs of the world in the 21st century,” notes a recent U.N. report on the global debt problem. “Developing countries must not be forced to choose between servicing their debt or serving their people,” and a reform of international financial architecture “is not only necessary, it is urgent.” Kenya’s upheaval this week is a direct illustration: The government’s effort to fund its budget with increased taxes on essential goods was a breaking point for many Kenyans, especially young, educated and unemployed youth.

Corruption has also slowed Kenya’s economic growth, diverting funds away from productive economic activities. Corruption is pervasive at almost every level of society, from police officers on the street to public procurement practices in government ministries. In the construction of Kenya’s $4.7 billion railroad, there were multiple reports of bribery, nepotism and extortion. Transparency International’s annual Corruption Perceptions Index currently ranks Kenya 126th of 180 countries. A report in March by the Office of the U.S. Trade Representative said corruption is a significant barrier to doing business in Kenya.

President Ruto campaigned for office two years on a pledge to reduce corruption, grow the economy and create jobs; he also promised to encourage major Western countries to reform the international financial architecture to better address the needs of developing countries. To steer Kenya away from more violent economic and political disruptions, he will need to make progress on both.

The views expressed in this publication are those of the author(s).

PUBLICATION TYPE: Analysis

V. I.   Lenin

The question of ministry of education policy [2], (supplement to the discussion on public education).

Written: Written April 27 (May 10), 1913 Published: First published in 1930 in the second and third editions of V. I. Lenin’s Collected Works , Vol. XVI. Published according to the manuscript. Source: Lenin Collected Works , Progress Publishers, 1977 , Moscow, Volume 19 , pages  137-146 . Translated: The Late George Hanna Transcription\Markup: R. Cymbala Public Domain: Lenin Internet Archive (2004). You may freely copy, distribute, display and perform this work; as well as make derivative and commercial works. Please credit “Marxists Internet Archive” as your source. • README

Our Ministry of Public (forgive the expression) “Education” boasts inordinately of the particularly rapid growth of its expenditure. In the explanatory note to the 1913 budget by the Prime Minister and the Minister of Finance we find a summary of the estimates of the Ministry of Public (so-called) Education for the post-revolutionary years. These estimates have increased from 46,000,000 rubles in 1907 to 137,000,000 in 1913. A tremendous growth—almost trebled in something like six years!

But our official praise-mongers who laud the police “law and order” or disorder in Russia ought not to have forgotten that ridiculously small figures always do grow with “ tremendous ” rapidity when increases in them are given as percentages. If you give five kopeks to a beggar who owns only three his “property” will immediately show a “tremendous” growth—it will be 167 per cent greater!

Would it not have been more fitting for the Ministry, if it did not aim at befogging the minds of the people and concealing the beggarly position of public education in Russia, to cite other data ? Would it not have been more fitting to cite figures that do not compare today’s five kopeks with yesterday’s three, but compare what we have with what is essential to a civilised state? He who does not wish to deceive either himself or the people should admit that the Ministry was in duty bound to produce these figures, and that by not producing such figures the Ministry was not doing its duty. Instead of making clear to the people, and the people’s representatives, what the needs of the state are, the   Ministry conceals these needs and engages in a foolish governmental game of figures, a governmental rehash of old figures that explain nothing.

I do not have at my disposal, of course, even a hundredth part of the means and sources for studying public education that are available to the Ministry. But I have made an attempt to obtain at least a little source material. And I assert boldly that I can cite indisputable official figures that really do make clear the situation in our official public “miseducation”.

I take the official government Russian Yearbook for 1910, published by the Ministry of the Interior (St. Petersburg, 1911).

On page 211, I read that the total number attending schools in the Russian Empire, lumping together primary, secondary and higher schools and educational establishments of all kinds, was 6,200,172 in 1904 and 7,095,351 in 1908. An obvious increase. The year 1905 , the year of the great awakening of the masses of the people in Russia, the year of the great struggle of the people for freedom under the leadership of the proletariat, was a year that forced even our hidebound Ministry to make a move.

But just look at the poverty we are doomed to, thanks to the retention of officialdom, thanks to the almighty power of the feudal landowners, even under conditions of the most rapid “departmental” progress.

The same Russian Yearbook relates in the same place that there were 46.7 people attending school to every 1,000 in habitants in 1908 (in 1904 the figure was 44.3 to every 1,000 inhabitants).

What do we learn from these figures from a Ministry of the Interior publication that the Ministry of Public Education did not feel inclined to report to the Duma? What does that proportion mean—less than 50 people out of a 1,000 attending school ?

It tells us, you gentlemen who uphold our hidebound public miseducation, of the unbelievable backwardness and barbarity of Russia thanks to the omnipotence of the feudal landowners in our state. The number of children and adolescents of school age in Russia amounts to over 20 per cent of the population, that is, to more than one-fifth . Even   Messrs. Kasso and Kokovtsov could without difficulty have learned these figures from their departmental clerks.

And so, we have 22 per cent of the population of school age and 4.7 per cent attending school, which is only a little more than one-fifth ! This means that about four-fifths of the children and adolescents of Russia are deprived of public education!

There is no other country so barbarous and in which the masses of the people are robbed to such an extent of education, light and knowledge—no other such country has remained in Europe; Russia is the exception. This reversion of the masses of the people, especially the peasantry, to savagery, is not fortuitous, it is inevitable under the yoke of the landowners, who have seized tens and more tens of millions of dessiatines of land, who have, seized state power both in the Duma and in the Council of State, and not only in these institutions, which are relatively low-ranking institutions....

Four-fifths of the rising generation are doomed to illiteracy by the feudal state system of Russia. This stultifying of the people by the feudal authorities has its correlative in the country’s illiteracy. The same government Russian Year book estimates (on page 88) that only 21 per cent of the population of Russia are literate, and even if children of pre -school age (i.e., children under nine) are deducted from the total population, the number will still be only 27 per cent.

In civilised countries there are no illiterates at all (as in Sweden or Denmark), or a mere one or two per cent (as in Switzerland or Germany). Even backward Austria-Hungary has provided her Slav population with conditions incomparably more civilised than feudal Russia has; in Austria there are 39 per cent of illiterates and in Hungary 50 per cent. It would be as well for our chauvinists, Rights, nationalists and Octobrists to think about these figures, if they have not set themselves the “statesmanlike” aim of forgetting how to think, and of teaching the same to the people. But even if they have forgotten, the people of Russia are learning more and more to think, and to think, furthermore about which class it is that by its dominance in the state condemns the Russian peasants to material and spiritual poverty.

America is not among the advanced countries as far as the number of literates is concerned. There are about 11 per cent   illiterates and among the Negroes the figure is as high as 44 per cent. But the American Negroes are more than twice as well off in respect of public education as the Russian peasantry. The American Negroes, no matter how much they may be, to the shame of the American Republic, oppressed, are better off than the Russian peasants—and they are better off because exactly half a century ago the people routed the American slave-owners, crushed that serpent and completely swept away slavery and the slave-owning state system, and the political privileges of the slave-owners in America.

The Kassos, Kokovtsovs and Maklakovs will teach the Russian people to copy the American example.

In 1908 there were 17,000,000 attending school in America, that is, 192 per 1,000 inhabitants — more than four times the number in Russia. Forty-three years ago, in 1870, when America had only just begun to build her free way of life after purging the country of the diehards of slavery—forty-three years ago there were in America 6,871,522 people at tending school, i.e., more than in Russia in 1904 and almost as many as in 1908. But even as far back as 1870 there were 178 ( one hundred and seventy-eight ) people enrolled in schools to every 1,000 inhabitants, little short of four times the number enrolled in Russia today .

And there, gentlemen, you have further proof that Russia still has to win for herself in persistent revolutionary struggle by the people that freedom the Americans won for them selves half a century ago.

The estimate for the Russian Ministry of Public Miseducation is fixed at 136,700,000 rubles for 1913. This amounts to only 80 kopeks per head of the population (170,000,000 in 1913). Even if we accept the “sum-total of state expenditure on education” that the Minister of Finance gives us on page 109 of his explanatory text to the budget, that is, 204,900,000 rubles, we still have only 1 ruble 20 kopeks per head. In Belgium, Britain and Germany the amount expended on education is two to three rubles and even three rubles fifty kopeks per head of the population. In 1910, America expended 426,000,000 dollars, i.e., 852,000,000 rubles or 9 rubles 24 kopeks per head of the population, on public education. Forty-three years ago,   in 1870, the American Republic was spending 126,000,000 rubles a year on education, i.e., 3 rubles 30 kopeks per head.

The official pens of government officials and the officials themselves will object and tell us that Russia is poor, that she has no money. That is true, Russia is not only poor, she is a beggar when it comes to public education. To make up for it, Russia is very “rich” when it comes to expenditure on the feudal state, ruled by landowners, or expenditure on the police, the army, on rents and on salaries of ten thousand rubles for landowners who have reached “high” government posts, expenditure on risky adventures and plunder, yesterday in Korea or on the River Yalu, today in Mongolia or in Turkish Armenia. Russia will always remain poor and beggarly in respect of expenditure on public education until the public educates itself sufficiently to cast off the yoke of feudal landowners.

Russia is poor when it comes to the salaries of school teachers. They are paid a miserable pittance. School-teachers starve and freeze in unheated huts that are scarcely fit for human habitation. School-teachers live together with the cattle that the peasants take into their huts in winter. School-teachers are persecuted by every police sergeant, by every village adherent of the Black Hundreds, by volunteer spies or detectives, to say nothing of the hole-picking and persecution by higher officials. Russia is too poor to pay a decent salary to honest workers in the field of public education, but Russia is rich enough to waste millions and tens of millions on aristocratic parasites, on military ad ventures and on hand-outs to owners of sugar refineries, oil kings and so on.

There is one other figure, the last one taken from American life, gentlemen, that will show the peoples oppressed by the Russian landowners and their government, how the people live who have been able to achieve freedom through a revolutionary struggle. In 1870, in America there were 200,515 school-teachers with a total salary of 37,800,000 dollars, i.e., an average of 189 dollars or 377 rubles per teacher per annum. And that was forty years ago! In America today there are 523,210 school-teachers and their total salaries come to 253,900,000 dollars, i.e., 483 dollars or 966 rubles per teacher per annum. And in Russia, even at the present   level of the productive forces, it would be quite possible at this very moment to guarantee a no less satisfactory salary to an army of school-teachers who are helping to lift the people out of their ignorance, darkness and oppression, if ... if the whole state system of Russia, from top to bottom, were reorganised on lines as democratic as the American system.

Either poverty and barbarism arising out of the full power of the feudal landowners, arising out of the law and order or disorder of the June Third law, or freedom and civilisation arising out of the ability and determination to win freedom—such is the object-lesson Russian citizens are taught by the estimates put forward by the Ministry of Public Education.

So far I have touched upon the purely material, or even financial, aspect of the matter. Incomparably more melancholy or, rather, more disgusting, is the picture of spiritual bondage, humiliation, suppression and lack of rights of the teachers and those they teach in Russia. The whole activity of the Ministry of Public Education in this field is pure mockery of the rights of citizens, mockery of the people. Police surveillance, police violence, police interference with the education of the people in general and of workers in particular, police destruction of whatever the people them selves do for their own enlightenment—this is what the entire activity of the Ministry amounts to, the Ministry whose estimate will be approved by the landowning gentry, from Rights to Octobrists inclusive.

And in order to prove the correctness of my words, gentlemen of the Fourth Duma, I will call a witness that even you, the landowners, cannot object to. My witness is the Octobrist Mr. Klyuzhev , member of the Third and Fourth Dumas, member of the supervisory council of the Second and Third Women’s Gymnasia in Samara, member of the school committee of the Samara City Council, member of the auditing board of the Samara Gubernia Zemstvo, former inspector of public schools. I have given you a list of the offices and titles (using the official reference book of the Third Duma) of this Octobrist to prove to you that the government itself , the landowners themselves in our landowners’ Zemstvo, have given Mr. Klyuzhev most important posts in   the “work” (the work of spies and butchers) of our Ministry of Public Stultification.

Mr.  Klyuzhev, if anybody, has, of course, made his entire career as a law-abiding, God-fearing civil servant. And, of course, Mr. Klyuzhev, if anybody, has by his faithful service in the district earned the confidence of the nobility and the landowners.

And now here are some passages from a speech by this most thoroughly reliable (from the feudal point of view) witness; the speech was made in the Third Duma in respect of the estimate submitted by the Ministry of Public Education.

The Samara Zemstvo, Mr. Klyuzhev told the Third Duma, unanimously adopted the proposal of Mr. Klyuzhev to make application for the conversion of some village two-year schools into four-year schools. The regional supervisor, so the law-abiding and God-fearing Mr. Klyuzhev reports, refused this. Why? The official explanation was: “ in view of the insignificant number of children of school age .

And so Mr. Klyuzhev made the following comparison: we (he says of landowner-oppressed Russia) have not a single four-year school for the 6,000 inhabitants of the Samara villages. In the town of Serdobol (Finland) with 2,800 in habitants there are four secondary (and higher than secondary) schools.

This comparison was made by the Octobrist, the most worthy Peredonov [1] ... excuse the slip, the most worthy Mr. Klyuzhev in the Third Duma. Ponder over that comparison, Messrs. Duma representatives, if not of the people, then at least of the landowners. Who made application to open schools? Could it be the Lefts? The muzhiks? The workers? God forbid! It was the Samara Zemstvo that made the application unanimously , that is, it was the Samara landowners , the most ardent Black-Hundred adherents among them. And the government, through its supervisor, refused the request on the excuse that there was an “ insignificant ” number of children of school age! Was I not in every way right when I said that the government hinders public education in Russia, that the government is the biggest enemy of public education in Russia?

The culture, civilisation, freedom, literacy, educated women and so on that we see in Finland derive exclusively from there being no such “social evil” as the Russian Government in Finland, Now you want to foist this evil on Fin land and make her, too, an enslaved country. You will not, succeed in that, gentlemen! By your attempts to impose political slavery on Finland you will only accelerate the awakening of the peoples of Russia from political slavery!

I will quote another passage from the Octobrist witness, Mr. Klyuzhev. “How are teachers recruited?” Mr. Klyuzhev asked in his speech and himself provided the following answer:

“ One prominent Samara man, by the name of Popov, bequeathed the necessary sum to endow a Teachers’ Seminary for Women.” And who do you think was appointed head of the Seminary? This is what the executor of the late Popov writes: “ The widow of a General of the Guard , was appointed head of the Seminary and she herself admitted that this was the first time in her life she had heard of the existence of an educational establishment called a Teachers’ Seminary for Women”!

Don’t imagine that I took this from a collection of Demyan Bedny’s fables, from the sort of fable for which the magazine Prosveshcheniye was fined and its editor imprisoned. Nothing of the sort. This fact was taken from the speech of the Octobrist Klyuzhev, who fears (as a God-fearing and police-fearing man) even to ponder the significance of this fact. For this fact, once again, shows beyond all doubt that there is no more vicious, no more implacable enemy of the education of the people in Russia than the Russian Government. And gentlemen who bequeath money for public education should realise that they are throwing it away, worse than throwing it away. They desire to bequeath their money to provide education for the people, but actually it turns out that they are giving it to Generals of the Guards and their widows . If such philanthropists do not wish to throw their money away they must understand that they should bequeath it to the Social-Democrats, who alone are able to use that money to provide the people with real education that is really independent of “Generals of the Guards”—and of timorous and law-abiding Klyuzhevs.

Still another passage from the speech of the same Mr. Klyuzhev.

“ It was in vain that we of the Third Duma desired free access to higher educational establishments for seminar pupils. The Ministry did not deem it possible to accede to our wishes.” “Incidentally the government bars the way to higher education, not to seminar pupils alone, but to the children of the peasant and urban petty-bourgeois social estates in general. This is no elegant phrase but the truth,” exclaimed the Octobrist official of the Ministry of Public Education. “Out of the 119,000 Gymnasium students only 18,000 are peasants. Peasants constitute only 15 per cent of those studying in all the establishments of the Ministry of Public Education. In the Theological Seminaries only 1,300 of the 20,500 pupils are peasants. Peasants are not admitted at all to the Cadet Corps and similar institutions.” (These passages from Klyuzhev’s speech were, incidentally, cited in an article by K. Dobroserdov in Nevskaya Zvezda No. 6, for 1912, dated May 22, 1912.)

That is how Mr. Klyuzhev spoke in the Third Duma. The depositions of that witness will not be refuted by those who rule the roost in the Fourth Duma. The witness, against his own will and despite his wishes, fully corroborates the revolutionary appraisal of the present situation in Russia in general, and of public education in particular. And what, indeed, does a government deserve that, in the words of a prominent government official and member of the ruling party of Octobrists, bars the way to education for the peasants and urban petty bourgeois?

Imagine, gentlemen, what such a government deserves from the point of view of the urban petty bourgeoisie and the peasants!

And do not forget that in Russia the peasants and the urban petty bourgeoisie constitute 88 per cent of the population, that is, a little less than nine-tenths of the people. The nobility constitute only one and a half per cent . And so the government is taking money from nine-tenths of the people for schools and educational establishments of all kinds and using that money to teach the nobility, barring the way to the peasant and urban petty bourgeois! Is it not clear what this government of the nobility deserves? This government that oppresses nine-tenths of the population in order to preserve the privileges of one-hundredth of the population—what does it deserve?

And now, finally, for the last quotation from my witness, the Octobrist official of the Ministry of Public Education, and member of the Third (and Fourth) Dumas, Mr. Klyuzhev:

“ In the five years from 1906 to 1910,” said Mr. Klyuzhev, “in the Kazan area, the following have been removed from their posts: 21 head masters of secondary and primary schools, 32 inspectors of public schools and 1,054 urban school-teachers; 870 people of these categories have been transferred. Imagine it,” exclaimed Mr. Klyuzhev, “how can our school-teacher sleep peacefully? He may go to bed in Astrakhan and not be sure that he will not be in Vyatka the next day. Try to understand the psychology of the pedagogue who is driven about like a hunted rabbit!”

This is not the exclamation of some “Left” school-teacher, but of an Octobrist. These figures were cited by a diligent civil servant. He is your witness, gentlemen of the Right, nationalists and Octobrists! This witness of “yours” is compelled to admit the most scandalous, most shameless and most disgusting arbitrariness on the part of the government in its attitude to teachers! This witness of yours , gentlemen who rule the roost in the Fourth Duma and the Council of State, has been forced to admit the fact that teachers in Russia are “ driven ” like rabbits by the Russian Government!

On the basis provided by this fact, one of thousands and thousands of similar facts in Russian life, we ask the Russian people and all the peoples of Russia: do we need a government to protect the privileges of the nobility and to “ drive ” the people’s teachers “like rabbits”? Does not this government deserve to be driven out by the people?

Yes, the Russian people’s teachers are driven like rabbits. Yes, the government bars the way to education to nine-tenths of the population of Russia. Yes, our Ministry of Public Education is a ministry of police espionage, a minis try that derides youth, and jeers at the people’s thirst for knowledge. But far from all the Russian peasants, not to mention the Russian workers, resemble rabbits , honourable members of the Fourth Duma. The working class were able to prove this in 1905, and they will be able to prove again, and to prove more impressively, and much more seriously, that they are capable of a revolutionary struggle for real freedom and for real public education and not that of Kasso or of the nobility.

[1] Peredonov—a type of teacher-spy and dull lout from Sologub’s novel The Petty Imp . — Lenin

[2] Lenin prepared this draft speech for a Bolshevik deputy to the Duma; the speech was delivered on June 4 (17), 1913 by A. E. Badayev during the debate on the Budget Committee’s report on estimates of the Ministry of Education for 1913. The greater part of Lenin’s draft was read almost word for word by Badayev, but he did not finish the speech. When he read the sentence “Does not this government deserve to be driven out by the people?” he was deprived of the right to speak.

Ending Learning Poverty and Building Skills: Investing in Education from Early Childhood to Lifelong Learning

The World Bank is the largest external financier of education in the developing world. We support education programs in more than 100 countries and are committed to helping countries increase access to quality education at all levels, reduce Learning Poverty, and develop skills, by putting in place education systems that assure opportunities for all.