I  The Human Capital Index 2020 UPDATE Human Capital in the Time of COVID-19 © 2020 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. The Human Capital Index 2020 UPDATE Human Capital in the Time of COVID-19 IV Acknowledge ments Acknowledgements The Human Capital Index is a collaboration between the Human Development Practice Group and the Development Economics Group of the World Bank. The 2020 update was led by Roberta Gatti and Aart Kraay and produced by Paul Corral, Nicola Dehnen, Ritika D’Souza and Juan Mejalenko. Noam Angrist, Syedah Aroob Iqbal, and Harry Patrinos updated the Harmonized Test Score outcomes. We are grateful to Pablo Ariel Acosta, Rita Kullberg Almeida, D. H. C. Aturupane, Anne Margreth Bakilana, Tekabe Ayalew Belay, Paolo Belli, Livia M. Benavides, Kamel Braham, Fadila Caillaud, Carine Clert, Jorge Coarasa, Gabriel Demombynes, Heba Elgazzar, Sameh El-Saharty, Stefan Emblad, Lire Ersado, Antonio Giuffrida, Inaam Ul Haq, Susanna Hayrapetyan, Samira Ahmed Hillis, Camilla Holmemo, Keiko Inoue, Timothy Johnston, Pierre Joseph Kamano, Olga Khan, Christophe Lemiere, Yasuhiko Matsuda, Muna Meky, Sophie Naudeau, Dorota Agata Nowak, Emre Ozaltin, Aleksandra Posarac, Maria Laura Sanchez Puerta, Hnin Hnin Pyne, Jamele P. Rigolini, Rafael Rofman, Cristina Isabel Panasco Santos, Aparnaa Somanathan, Lars Sondergaard, Michel Welmond, William Wiseman, Ruslan Yemtsov and Xiaoqing Yu for careful data review. This report was written by a core team led by Roberta Gatti and including Paul Corral, Nicola Dehnen, Ritika D’Souza and Juan Mejalenko. Steven Pennings wrote the chapter on Human Capital Utilization. This report benefitted from Aart Kraay’s advice and from analytical inputs by: Daniel Halim (gender analysis), Amer Hasan and Fiona Mackintosh (case studies narrative), Jigyasa Sharma (on fragility), Joao Pedro de Azevedo and Diana Goldemberg (COVID-19 impact on learning-adjusted years of schooling), Dina Abu-Ghaida and Mohamed Audah (schooling in Syria), Alejandro de la Fuente (schooling in Sierra Leone), Chloé Desjonquères (learning progress in Ceará), Alina Sava and Lars Sondergaard (schooling in Romania), Utz Pape (Rapid Response Phone Surveys), Halsey Rogers (Challenges in test-score com- parison over time), Saskia de Pee and Cecilia Garzón (World Food Program) and Naveed Akbar (Benazir Income Support Programme, Government of Pakistan) and Emanuela Galasso, Lisa Saldanha, Meera Shekar, Marie-Chantal Uwanyiligira, and Kavita Watsa (cross-sectoral approaches to stunting). We are indebted to David Weil for his overarching guidance. We are grateful to our peer reviewers Shubham Chaudhuri, Rachel Glennerster, William Maloney, and David Weil for their insightful views and to Deon Filmer for his detailed comments on earlier versions of this draft. We thank Kathleen Beegle, Hana Brixi, Emanuela Galasso, Ramesh Govindaraj, Ambar Narayan, Meera Shekar, Sharad Tandon, Tara Vishwanath, and Michael Weber for thoughtful comments and conversations. We are grateful to Alex Irwin for his outstanding editing touch, to Chloé Desjonquères for efficiently managing the report’s production pro- cess, and to Ruben Conner, Mary Fisk, Sebastian Insfran, and Andres Yi Chang for their careful read of the report. Finally, we are also grateful to Luis Eduardo San Martin and Luiza Andrade from the DIME Analytics team for a thorough code review. This Human Capital Index update was developed under the strategic guidance of Mari Pangestu, Annette Dixon, and Mamta Murthi and benefitted from the views of Nadir Mohammed and Alberto Rodriguez. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 V Table of Contents Acknowledgements ............................................................................................................................ IV Executive Summary.............................................................................................................................. IX Overview............................................................................................................................................... XIII 1 The Human Capital Index 2020 Update ...................................................................... 1 1.1 The Human Capital Index methodology .......................................................................... 2 1.2 The Human Capital Index 2020........................................................................................... 4 1.3 HCI 2020 Update—index components ............................................................................ 11 1.4 HCI Measures of Gender Gaps in Human Capital ....................................................... 17 1.5 Human Capital in fragile and conflict-affected contexts............................................ 21 1.6 HCI 2020 Update table.......................................................................................................... 27 2 Human Capital Accumulation over Time................................................................... 30 2.1 Human capital accumulation over the Past decade...................................................... 31 2.2 Changes in key human capital dimensions in the past decade............................... 36 2.3 A longer-run view of country progress........................................................................... 56 3 Accumulation Interrupted? COVID-19 and Human Capital.................................... 62 3.1 Transmission of the COVID-19 shock to Human Capital......................................... 63 3.2 The COVID-19 Human Capital Shock: A Life-Cycle Perspective ......................... 66 3.3 Using the HCI to Simulate the Impact of the Pandemic .......................................... 74 3.4 Annex 3A: COVID-19 Shock to the Under-5 Cohorts................................................ 80 3.5 Annex 3B: COVID-19 shock to school age cohorts..................................................... 82 VI Tab le of C ontents 4 Utilizing Human Capital................................................................................................. 84 4.1 Methodology and the Basic UHCI Measure................................................................... 86 4.2 The Basic Utilization-adjusted HCI in the data............................................................. 87 4.3 The Full Utilization-adjusted HCI .................................................................................... 91 4.4 Full Utilization-adjusted HCI in the data....................................................................... 93 4.5 Comparing the Utilization Measures............................................................................... 94 4.6 Disaggregation by Region................................................................................................... 95 4.7 Disaggregation by Gender................................................................................................... 96 5 Informing policies to protect and build human capital ........................................ 102 5.1 Good measurement: necessity, not luxury................................................................... 103 5.2 Beyond the Human Capital Index ................................................................................. 104 5.3 Building, protecting, and employing human capital in a post pandemic world.... 109 5.4 A data-driven health sector response............................................................................. 109 5.5 Preventing losses in learning............................................................................................. 113 5.6 Reinforcing resilience among vulnerable people and communities................... 113 5.7 Coordinating action across sectors and adopting a whole-of-society approach.... 114 References ........................................................................................................................................... 116 Appendices......................................................................................................................................... 129 Appendix A: The Human Capital Index: Methodology............................................... 130 Appendix B: Back-calculated HCI................................................................................... 140 Appendix C: HCI Component Data Notes..................................................................... 144 T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 IX Executive Summary T he Human Capital Index (HCI) is an international metric that benchmarks key components of human capital across countries. Measuring the human capital that a child born today can expect to attain by her 18th birthday, the HCI highlights how current health and education outcomes shape the productivity of the next generation of workers. In this way, it underscores the importance for govern- ments and societies of investing in the human capital of their citizens. The HCI was launched in 2018 as part of the Human Capital Project (HCP), a global effort to accelerate progress towards a world where all children can achieve their full potential. Over the past decade, many countries have made important progress in improving human capital. Today, however, the COVID-19 pandemic threatens to reverse many of those gains. Urgent action is needed to protect hard-won advances in human capital, particularly among the poor vulnerable. Designing the needed interventions, targeting them to achieve the highest effectiveness, and navigating difficult trade-offs in times of reduced fiscal space, makes investing in better measurement of human capital more important than ever. Human capital consists of the knowledge, skills, and health that people accumulate over their lives. People’s health and education have undeniable intrinsic value, and human capital also enables people to realize their potential as productive members of society. More human capital is associated with higher earnings for people, higher income for countries, and stronger cohesion in societies. It is a central driver of sustainable growth and poverty reduction. This report accompanies the release of 2020 data on the HCI. Building on momentum from the first edi- tion in 2018, the 2020 issue updates the index using new and expanded data for each of the HCI compo- nents through March 2020. As such, the report provides a snapshot of the state of human capital before COVID-19 and a baseline to track the pandemic’s impacts on human capital. COVID-19 struck at a time when the world was healthier and more educated than ever. Yet, data pre- sented in this report reveal that substantial human-capital shortfalls and equity gaps existed before the crisis. Worldwide, a child born just before the advent of COVID-19 could expect to achieve on average just 56 percent of her potential productivity as a future worker. Gaps in human capital remain especially deep in low-income countries and those affected by violence, armed conflict, and institutional fragility. Expanded sex-disaggregated data show that girls currently enjoy a slight edge over boys in human capital accumulation in most countries, reflecting in part a female biological advantage early in life. However women continue to be at a substantial disadvantage in many dimensions of human capital that are not captured by the HCI’s components, including participation in economic life. X Executive S u mmary In addition to describing HCI data and methodology, this report documents the evolution of human capital over the last decade. Human capital outcomes progressed in almost all countries by about 4 per- cent on average during this period, thanks primarily to better health and increased access to schooling. However, many countries struggled to improve learning outcomes, as educational quality often failed to keep pace with gains in enrollment. The various dimensions of human capital improved with economic development, and they did so at a surprisingly similar pace across country income groups. Progress was only slightly faster in low-income countries, which are further away from the frontier of full health and education. The trajectories of individual countries differed considerably, including in how human-capital gains were distributed across the socio-economic spectrum within each country. In some contexts, the most dis- advantaged groups scored the greatest gains. In others, poorer and richer families benefitted equally. Along with broad economic development, specific policies contributed to some countries’ progress in human capital. Effective policies included expanding the population coverage of health services, notably for maternal and child health; bolstering nutrition and access to sanitation; making school more afford- able; and providing financial support to vulnerable families through mechanisms such as cash transfer programs and insurance. Strong gains were more likely in countries that were able to maintain commit- ment to reform across political cycles and to adopt an evidence-based, whole-of-society approach to policymaking. These same elements will be essential to protect human capital in the face of the COVID-19 crisis. While data on COVID-19’s impacts on human-capital outcomes are only beginning to emerge, simulations conducted for this report suggest that school closures combined with family hardship are significantly affecting the accumulation of human capital for the current generation of school-age children. The impacts appear comparable in magnitude to the gains that many countries achieved during the previous decade, suggesting that the pandemic may roll back many years’ worth of human-capital progress. In parallel, COVID-19’s disruption of health services, losses in income, and worsened nutrition are expected to increase child mortality and stunting, with effects that will be felt for decades to come. The HCI can be a useful tool to track such losses and guide policy to counter them, since the index is based on robust markers for key stages of human-capital accumulation in the growth trajectory of a child. However, the five components of the HCI do not cover all the important aspects of the accumulation and productive use of human capital. In particular, the index is silent on the opportunities to use accumulated human cap- ital in adulthood through meaningful work. In many countries, a sizable fraction of today’s young people may not be employed when they become adults. Even if they find employment, they may not hold jobs where they can use their skills and cognitive abilities to increase their productivity. Recognizing the salience of such patterns for how human capital gains are translated into economic progress and shared prosperity, this report analyzes two measures that augment the HCI to account for the utilization of human capital. These measures provide insight on further margins that countries can explore to boost their long-term growth and productivity. Both utilization measures suggest that human capital is particularly underutilized in middle-income countries. A key message is that human capital is also strikingly underutilized for women in many settings: the gender gap in employment rates (a basic measure of utilization) is 20 percentage points on average worldwide, but exceeds 40 percentage points in South Asia and the Middle East and North Africa. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 XI By bringing salience to the productivity implications of shortfalls in health and education, the HCI has not only clarified the importance of investing in human capital, but also highlighted the role that mea- surement can play in catalyzing consensus for reform. Better measurement enables policy makers to design effective interventions and target support to those who are most in need, which is often where interventions yield the highest payoffs. Investing in better measurement and data use now is a neces- sity, not a luxury. In the immediate, it will guide pandemic containment strategies and support for the most affected. In the medium term, better curation and use of administrative, survey, and identification data will be essential to guide policy choices in an environment of limited fiscal space and competing priorities. Today, hard-won human capital gains in many countries are at risk. But countries can do more than just work to recover the lost ground. Ambitious, evidence-driven policy measures in health, education, and social protection can pave the way for today’s children to surpass the human-capital achievements and quality of life of the generations that preceded them. To protect and extend earlier human-capital gains, policymakers need to expand health service cov- erage and quality among marginalized communities, boost learning outcomes together with school enrollments, and support vulnerable families with social protection measures adapted to the scale of the COVID-19 crisis. Informed by rigorous measurement, bold policies can drive a resilient recovery from the pandemic and open a future in which rising generations will be able to develop their full potential and use it to tackle the vast challenges that still lie ahead for countries and the world: from ending pov- erty to preventing armed conflict to controlling climate change. COVID-19 has underscored the shared vulnerability and common responsibility that today link all nations. Fully realizing the creative promise embodied in each child has never been more important. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 XIII Overview T he Human Capital Index (HCI) measures the human capital that a child born today can expect to attain by her 18th birthday, given the risks of poor health and poor education prevailing in her country.1 The index incorporates measures of different dimensions of human capital: health (child survival, stunting and adult survival rates) and the quantity and quality of schooling (expected years of schooling and international test scores). Human capital has intrinsic value that is undeniably important, but difficult to quantify. This in turn makes it challenging to combine its different components into a sin- gle measure. The HCI uses global estimates of the economic returns to education and health to create an integrated index that captures the expected productivity of a child born today as a future worker, relative to a benchmark – the same for all countries – of complete education and full health. THE HCI 2020 UPDATE The 2020 update of the HCI incorporates the most recent available data to report HCI scores for 174 countries, adding 17 new countries to the index relative to the 2018 edition. The 2020 update uses new and expanded data for each of the HCI components, available as of March 2020. As in 2018, data were obtained from official sources and underwent a careful process of review and curation. Given the timing of data collection, this update can serve as a benchmark of the levels of human capital accumulation that existed immediately prior to the onset of the COVID-19 pandemic. Globally, the HCI 2020 shows that, before the pandemic struck, a child could expect to attain an average of 56 percent of her potential productivity as a future worker. This global average masks considerable vari- ation across regions and economies. For instance, a child born in a low-income country could expect to be 37 percent as productive as if she had full education and full health. For a child born in a high-income country, this figure is 70 percent. INCOME ALONE DOES NOT EXPLAIN COUNTRY DIFFERENCES IN HUMAN CAPITAL What explains these variations in human capital outcomes? While the correlation between HCI and gross domestic product (GDP) per capita is strong, human capital does not always move in lockstep with economic development. Countries like Burundi, Estonia, Kyrgyz Republic, Uzbekistan, and Vietnam have outcomes 1 The HCI was introduced in World Bank (2018a, 2018b), and the methodology of the HCI is detailed in Kraay (2019). XIV Overview that are higher than predicted by their GDP per capita. Conversely, in a number of countries, human cap- ital is lower than per capita income would suggest. Among these are several resource-rich countries, where human capital development has not yet matched the potential that one would anticipate, given these coun- tries’ wealth. Differences in the quantity and quality of schooling account for the largest part of HCI differences across country-income groups. Of the 33 percentage-point difference between the scores of the average low- and high-income country, almost 25 percentage points are accounted for by the differences in learn- ing-adjusted years of school, a measure which combines expected years of school with learning as mea- sured by harmonized test scores (i.e. test scores that are made comparable across countries). While education drives HCI differences across country-income groups, education’s contribution to gaps within these groups varies by income level. For instance, education accounts for roughly 90 percent of the difference between high and low performers within high-income country groups, but only 60 per- cent within the group of low-income economies. In contrast, differences in child survival rates account for less of the difference in HCI scores among high-income countries, largely because child mortality is low across these countries. The same is true for health differences, which explain a lower share of country differences in the HCI as one moves from low- to higher-income groups, since health outcomes tend to be uniformly better as countries get richer.2 Human capital outcomes also vary for girls and boys. A disaggregation of the HCI by gender—now avail- able for 153 of the 174 included countries—shows that human capital is slightly higher among girls than boys in most countries. Girls are not only catching up to but outperforming boys in expected years of schooling and learning outcomes in some regions. For example, in the Middle East and North Africa, girls can expect to complete more than half of an additional learning-adjusted year of school compared with boys. However, the reverse is true in Sub-Saharan Africa and in South Asia. Investing in human capital enhances social cohesion and equity while strengthening people’s trust in institutions. Nowhere is this more important than in countries grappling with fragility and conflict. External shocks such as armed conflict and natural disasters have destructive impacts on both countries’ existing human capital stock and on the process of building new human capital. Evidence increasingly suggests that, for armed conflict as well as famine, these negative effects can persist for decades and even across generations. This weakens the core of sustainable and equitable economic development. Unfortunately, yet unsurprisingly, the HCI 2020 indicates that, on average, countries impacted by fra- gility, conflict, and violence have lower HCI values, compared to the rest of the world. In particular, the seven countries with the lowest HCI 2020 scores are also on the World Bank’s current list of fragile and conflict-affected situations (FCS). This adds to the urgency of addressing human-capital gaps in FCS set- tings. Only by preserving and rebuilding human capital can countries durably escape cycles of fragility and underdevelopment. 2 Stunting and adult survival are here considered together for easy comparison. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 XV MOST COUNTRIES ACHIEVED HUMAN CAPITAL GAINS IN THE DECADE BEFORE COVID-19 Since this is the first update of the HCI, the 2020 release presents an opportunity to assess the evolution of human capital outcomes as measured by the index over the last decade. The HCI is based on outcomes that typically change slowly from year to year. Some of them—such as stunting and educational test scores—are measured infrequently, every three to five years. As a result, changes in the HCI over a short period are small and may simply reflect updates to some components that are measured sporadically but not others. To provide more reliable insights on countries’ human capital trajectories over time, this report focuses on changes in the index over the past decade. To this end, a (circa) 2010 version of the HCI is constructed with data carefully curated to maximize comparability with the 2020 results. In particular, only those countries where learning scores were measured by the same international assessment program in 2010 and 2020 enter the comparison.3 The resulting sample for the 2010 HCI includes 103 countries.4 As measured by the HCI, human capital progressed in the vast majority of countries in this sample. On average, between 2010 and 2020, the HCI improved by 2.6 percentage points, about 4 percent of its aver- age value in 2010.5 One economy in four that experienced a rise in the index recorded gains of more than five percentage points. This means that, in those countries, the productivity of future workers approached the frontier of full productivity by five percentage points – a substantial achievement. Economies starting from lower levels of human capital improved by larger amounts. Better health (child and adult survival and stunting) accounts for about half of the HCI’s changes. Increased enrollments—especially at pre-pri- mary and secondary school levels—account for the bulk of remaining changes. In contrast, progress on learning outcomes has proved difficult, as international test scores failed to keep pace with enrollment gains in many settings. In the human capital dimensions captured by the index, girls and boys made similar progress over time, with only a handful of countries reporting opposite trends. In the 90 countries where disaggregated data are available and comparisons with 2010 are possible, the average gender ratio is similar in 2010 and 2020, at about 1.06 in favor of girls. Around 2010, the HCI was uniformly larger among girls than among boys, with the exception of seven economies. Among these, the girl-boy ratio improved, approaching or surpassing gender parity, in all but of one country over the past decade. DATA FOR SELECTED COUNTRIES SHOW HOW DISADVANTAGED HOUSEHOLDS SHARED IN HUMAN CAPITAL GAINS National averages also mask differential trends in human capital between richer and poorer households. Using household data from Demographic and Health Surveys and Multiple Indicator Cluster Surveys, it is possible to calculate a version of the HCI disaggregated by socioeconomic status (SES) for a number of 3 As described in chapter 2, this rule is relaxed only for five countries in the sample. These countries are included in the sample with learning scores from different international assessments (TIMMS/PIRLS in 2010 and PISA in 2020). To increase comparability, only scores for secondary schooling are considered for 2010. 4 This sample is, unsurprisingly, skewed toward richer countries where data tend to be more complete and of better quality. 5 With the sample of countries on which the comparison is based skewed towards richer countries, which are closer to the frontier and would naturally have slower change in their human capital, the pace of change is likely underestimated. XVI Overview low-and middle-income countries. Countries vary substantially in how gains in human-capital outcomes are distributed across the population.6 For instance, Haiti, Malawi, and Senegal all improved their child survival rates over the last decade. However, the gap between rich and poor households in Haiti remained constant, while it decreased in Malawi and Senegal.7 Similarly, the years of schooling a child could expect in Burkina Faso, Bangladesh, and India increased significantly. But in Burkina Faso the six-year gap in Expected Years of Schooling (EYS) between rich and poor households has stayed constant over the past 10 years, while in the same period Bangladesh and India—albeit starting from different levels—were able to halve the gap between their richest and poorest households. Côte d’Ivoire’s 25 percentage-point gap between stunting rates for rich and poor households remained unchanged, notwithstanding a signif- icant average reduction in stunting. Conversely, Uganda was able to narrow this gap from a difference of 20 to 16 percentage points between 2000 and 2016. Addressing such rich-poor gaps in human capital must remain a priority for governments committed to equitable growth, not least because the returns to investment in human capital are often highest for disadvantaged groups, especially for measures that act early in life. Human capital is a central driver of sustainable growth and poverty reduction. However, even for gov- ernments that recognize the importance of investing in the human capital of their citizens, the process of designing policy and building institutions that foster human capital accumulation can be complex, with the full benefits taking years and even decades to materialize. This is evidenced in the relatively modest progress measured for the average country on the HCI over the last decade. Adopting a longer timeframe can help identify many forms of government action that can improve human capital. For that purpose, this report incorporates insights from case studies to better understand the trajectories of countries that have made notable improvements in various dimensions of human capital. Sustained political commit- ment spanning election cycles; coordination across the many programs and agencies that may influence human capital; and using a robust evidence base to inform policy choices emerge as key elements con- tributing to successful policies for human capital.8 SIMULATIONS USING THE HCI QUANTIFY COVID-19’S IMPACTS ON HUMAN CAPITAL COVID-19 is placing countries’ hard-won human capital gains at risk. A lesson from past pandemics and crises is that their effects are not only felt by those directly impacted, but often ripple across populations and, in many cases, across generations. This underscores the urgent need to protect and rebuild human capital to foster recovery in the short and longer terms. Setbacks during certain life stages - chiefly early childhood - can have especially damaging and long-last- ing effects on human capital accumulation. For example, during childhood, the link between parental income and child health is particularly strong.9 In previous crises, poorer nutrition and reduced well- being among pregnant mothers led to permanent losses in their children’s cognitive attainment, as 6 The analysis of SES-disaggregated HCI outcomes is based on D’Souza, Gatti, and Kraay (2019). 7 It is important to note the dramatic increase in child mortality that occurred in Haiti in 2010 in the aftermath of the country’s catastrophic January 2010 earthquake. 8 This approach informs the work of the World Bank’s Human Capital Project (HCP). 9 See Almond (2006). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 XVII well as higher chronic disease rates when the children became adults.10 COVID-19 is likely to produce similar outcomes. In this crisis, human-capital impacts associated with economic shocks come atop reductions in care linked to service disruptions during the pandemic’s acute phase. As such, the current shock to family incomes, even if transitory, may have repercussions for years to come. Children in dis- advantaged families will be disproportionately vulnerable to all these effects, thus deepening existing inequalities. The HCI methodology can be used to quantify some of the potential impacts of COVID-19 on the future human capital of children and youth. For young children—those born during the pandemic or who are currently under the age of five—disruptions to health systems, reduced access to care, and family income losses will materialize as increased child mortality, malnutrition, and stunting. Because stunting and edu- cational outcomes are closely intertwined, the pandemic risks durably setting back children’s learning. According to HCI-based simulations, in low-income countries, young children today can expect their human capital to be about 1 percent lower than it would have been in the absence of COVID-19. At the height of the pandemic, close to 1.6 billion children worldwide were out of school. For most chil- dren who are currently of school age, the pandemic has meant that formal teaching and learning no longer happen face to face. Since the ability to roll out distance learning differ across countries, and even within countries, considerable losses in schooling and learning can be anticipated. The income shocks associated with COVID-19 will also force many children to drop out of school. Putting these effects together suggests that the pandemic could reduce global average learning-adjusted years of school by half a year, from 7.8 to 7.3 years. Translated into the terms of the HCI itself, this loss means a drop of almost 4.5 percent in the HCI of the current cohort of children. For a country with an HCI of 0.5, this signifies a drop of 0.025 HCI points, a reduction of the same order of magnitude as the HCI increase that many countries have achieved over the past decade. Without a strong policy response now, the pandemic’s negative human-capital effects will likely continue to reduce countries’ productivity and growth prospects for decades. In 20 years, roughly 46 percent of the workforce in a typical country (people aged 20 to 65 years) will be composed of individuals who were either in school or under the age of five during the COVID-19 pandemic. A typical country at that time could still show a loss in its HCI of almost 1 full HCI point (0.01) due to COVID-19. That is, even if the pandemic turns out to be a temporary shock, the COVID-19 shock could still leave current cohorts of children behind for the rest of their lives. No society can afford to let that happen. THIS UPDATE AUGMENTS THE HCI TO REFLECT HOW WELL HUMAN CAPITAL IS USED The HCI can be harnessed to track future human-capital losses and guide policies to limit them. The HCI has this capability because it is a metric based on reasonably directly measured markers for key stages of human capital in the growth trajectory of a child. However, the five components of the index do not cover all the important aspects of the accumulation and productive use of human capital. When today’s child becomes a future worker, in many countries she may not be able to find a job, and even if she can, 10 See Almond and Currie (2011). XVIII Overview it might not be a job where she can fully use her skills and cognitive abilities to increase her productivity. In these cases, her human capital can be considered underutilized. Recognizing the importance of this pattern both for individual people and for policy, this report analyzes two simple extensions of the HCI that adjust the HCI for labor market underutilization of human capital. Both utilization-adjusted human capital indexes (UHCIs) can be calculated for more than 160 countries. Both have the same simple form—the HCI multiplied by a utilization rate—and represent the long-run income gains if a country moves to complete human capital and full utilization of that human capital.11 The UHCIs are meant to complement, not replace the HCI, given their different purposes. The two UHCIs take different approaches to measuring utilization. In the basic UHCI, utilization is mea- sured as the fraction of the working-age population that is employed. While this measure is simple and intuitive, it is not able to capture the fact that a large share of employment in developing countries is in jobs where workers may not be able to fully use their human capital to increase their productivity. The full UHCI adjusts for this by introducing the concept of “better employment”­—defined as non-agricultural employees, plus employers—which are the types of jobs that are common in high-productivity coun- tries. The full utilization rate depends on the fraction of a country’s working-age population in better employment. Countries with higher HCI scores also face larger utilization penalties if they show low rates of better employment, as they have more human capital to underutilize. While the different methodologies produce different scores for some individual countries, the basic and full measures yield broadly similar utilization rates across country-income groups and regions, and in general. Utilization rates average around 0.6, but they follow U-shaped curves when plotted against per capita income across countries, being lowest over a wider range of lower-middle-income countries. The analysis of underutilization suggests that moving to a world with complete human capital and complete utilization of that human capital, long-run per capita incomes could almost triple. Both UHCIs reveal starkly different gender gaps from those calculated using the HCI. While the HCI is roughly equal for boys and girls, with a slight advantage for girls on average, UHCIs are lower for females than males, driven by lower utilization rates. Basic utilization (employment) rates are 20 per- centage points lower for women than men in general, and with a gap of more than 40 percentage points in the Middle East and North Africa and South Asia. Female employment rates follow strongly U-shaped curves when plotted against countries’ levels of income, whereas male employment rates are much flatter, and with less dispersion across countries. The gender gap is also present in the full utilization rate, though it is smaller. These results suggest that, while gender gaps in human capital in childhood and adolescence have closed in the last two decades (especially for education), major chal- lenges remain to translate these gains into opportunities for women. 11 Specifically, long run GDP per capita is 1/UHCI times higher in a world with complete human capital and complete utilization than under the status quo. This is a generalization of the interpretation of the HCI. See Pennings (2020) for details. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 XIX BETTER MEASUREMENT ENABLES BETTER POLICY As the COVID-19 crisis continues to unfold, data and measurement are more vital than ever to shape gov- ernments’ response in the immediate and guide future policy choices towards (cost-) effective solutions. Better measurement and data use are investments that pay off, a consideration that is particularly import- ant now, as countries face dwindling fiscal space and many competing demands. By generating a shared understanding among diverse actors, measurement can shine a light on constraints that limit progress in human capital. In the same way, effective measurement can facilitate political con- sensus based on facts and help muster support for reforms. Measurement also enables policy makers to target support to those who are most in need, which is often where interventions yield the highest payoffs. As policy implementation moves forward, measurement provides feedback to guide course corrections. In the context of a pandemic, governments that use relevant data in real time are better able to monitor the evolution of disease transmission and continuously update containment strategies, while responding to the immediate and long-term effects of the economic crisis on households and communities. At all times, data are especially important in countries affected by fragility or conflict, though measurement is far more difficult in these settings. The HCI offers a high-level view of human capital across countries that can help to catalyze new con- versations with key stakeholders. At the same time, much greater depth in measurement and research is needed to better understand the dynamics of human capital accumulation, including across socio- economic groups and geography, and how policies can affect it. Some key measurement improvements – such as leveraging phone surveys and making better use of administrative data – can be achieved in the short term. Other improvements will demand a more sustained effort from countries and develop- ment partners. These longer-range efforts include rethinking the architecture of country data systems to connect different administrative data sources, and fielding surveys to better understand the needs and behavior of teachers and health providers. The COVID-19 crisis threatens gains in human capital that countries have achieved through decades of effort. A renewed, society-wide commitment is needed to protect human capital in the immediate and to remediate the looming losses in the longer run. Challenges range from crafting context-sensitive school re-opening protocols to deeper reforms that will promote children’s learning at all stages: starting from cognitive stimulation in the early years, then continuing to nurture relevant skills throughout child- hood and adolescence. Building blocks for success will include better-prepared teachers, better-managed schools, and incentives that are aligned across the many stakeholders in education reform. Support to households will be essential not only to buffer income losses but also to sustain the demand side of schooling and health care. Such support can come through cash transfers, but also interventions aimed at reconnecting workers to jobs. Strengthening disease surveillance and a renewed commitment to universal health coverage will be essential to build resilient health systems that offer affordable, quality care to all. Investments in water, sanitation, and – increasingly – digitalization are important complements to sustain human capital accumulation. Current deepening inequalities in human capital outcomes make it imperative to target interventions to children from the most disadvantaged families. This is the way to prevent setbacks where they risk generating the worst consequences for people’s lifetime trajectories. XX Overview With fiscal space shrinking as competing priorities multiply, policymakers face hard choices. Proven strategies include engaging the whole of society, identifying cross-sectoral synergies, and using data to select cost-effective interventions and track their effective implementation. These approaches will not make tough policy trade-offs painless. But they will enable leaders to choose the options that have the highest probability of success. Applying these tools, governments can go far toward protecting and rebuilding human capital in the wake of COVID-19. And that is not all. Strong, evidence-driven human capital investments now can do far more than restore what has been lost. Health, education, social pro- tection, and other policies informed by rigorous measurement can take countries’ human capital beyond the levels previously achieved, opening the way to a more prosperous and inclusive future. 1 The Human Capital Index 2020 UPDATE T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 1 A t the organization’s 2018 Annual Meetings, from gaps in human capital across countries, the the World Bank Group launched the HCI underscores the urgency of improving human Human Capital Project, an unprecedented capital outcomes for children today. global effort to support human capital development as a core element of countries’ overall strategies to In response to the call for governments to invest increase productivity and growth. The main objec- in the human capital of their citizens, 77 coun- tive of the project is rapid progress toward a world tries across the world are now part of the Human in which all children can achieve their full potential. Capital Project. These countries have affirmed For that to happen, children need to reach school building, protecting, and employing human cap- well-nourished and ready to learn, attain real learn- ital as a national priority and have been prioritiz- ing in the classroom, and enter the job market as ing investments in human capital and undertaking healthy, skilled, and productive adults. difficult reforms, sometimes in very challenging contexts. With a view to maintaining this momen- Central to this effort has been the Human Capital tum, the 2020 update of the HCI incorporates the Index (HCI), a cross-country metric measuring the most recent data to report HCI scores for 174 coun- human capital that a child born today can expect tries, adding 17 new countries to the index relative to attain by her 18th birthday, given the risks of to the 2018 edition. poor health and poor education prevailing in her country.1 The HCI brings together measures of dif- The update uses new and expanded data for ferent dimensions of human capital: health (child each of the HCI components, with a cut-off date survival, stunting, and adult survival rates) and the of March 2020. Computed using data collected quantity and quality of schooling (expected years before COVID-19 had impact on a global scale, the of school and international test scores). Using esti- HCI 2020 provides a useful benchmark to track mates of the economic returns to education and the evolution of human capital and its key compo- health, the components are combined into an nents in the wake of the pandemic. index that captures the expected productivity of a child born today as a future worker, relative to a Sections 2 and 3 of this chapter outline the HCI benchmark of complete education and full health. methodology and describe the main features of the HCI 2020 and its components. Section 4 discusses The Human Capital Index ranges from 0 to 1, so gender differences across countries and regions. that an HCI value of, for instance, 0.5 implies that a Finally, section 5 presents a special spotlight sec- child born today will only be half as productive as a tion that considers the unique human capital chal- future worker as she would be if she enjoyed com- lenges that arise in states grappling with fragility, plete education and full health. By benchmarking conflict, and violence. Section 6 reports the HCI shortfalls in future worker productivity deriving 2020 scores for 174 countries. 1 The HCI was introduced in World Bank (2018a, 2018d), and the methodology of the HCI is detailed in Kraay (2018). 2 The Hum an Capita l Index 20 20 U pdate 1.1 THE HUMAN CAPITAL INDEX action in the short to medium term. The need to METHODOLOGY produce such a metric has oriented the choice of components toward measuring the human capital The HCI is designed to highlight how improve- of the next generation, rather than measuring the ments in current health and education outcomes stock of human capital of the current workforce, shape the productivity of the next generation which largely reflects policy choices made decades of workers, assuming that children born today ago, when the current workforce was of school age.2 experience over the next 18 years the educational As a result, the HCI quantifies the key stages in a opportunities and health risks that children in this child’s human capital trajectory and their conse- age range currently face. quences for the productivity of the next generation of workers, with three components: The HCI captures key stages of a child’s trajectory from birth to adulthood. In the poorest countries in Component 1—Survival from birth to school age, the world, there is a significant risk that a child will measured using under-5 mortality rates. not survive to her fifth birthday. Even if she does reach school age, there is a further risk that she will Component 2—Expected years of learning-ad- not start school, let alone complete the full cycle of justed school, combining information on the 14 years of schooling, from preschool to grade 12, quantity and quality of education. The quantity of which is the norm in rich countries. The time she education is measured as the number of years of does spend in school may translate unevenly into school a child can expect to obtain by age 18 given learning, depending on a variety of factors includ- the prevailing pattern of enrollment rates across ing the quality of teachers and schools that she grades. The quality of education reflects work experiences. When she turns 18, she carries with undertaken at the World Bank to harmonize test her the lasting effects of poor health and nutrition scores from major international student achieve- during childhood that limit her physical and cog- ment testing programs (Patrinos and Angrist, 2018). nitive abilities as she develop into adulthood. These are combined into a measure of learning-ad- justed school years as proposed in the 2018 World The design of the HCI has been guided by a num- Development Report (see Box 1.1). ber of criteria. First, the HCI is outcome- rather than inputs- based. This helps focus the conversa- Component 3—Health. In the absence of a single tion on what matters—results—and provides incen- broadly-accepted, directly measured, and widely tives for countries not only to invest more, but also available metric, the overall health environment to invest better in human capital, without concerns is captured by two proxies: (a) adult survival rates, that the HCI might be susceptible to gaming. The defined as the fraction of 15-year-olds who survive likelihood that a cross-country benchmarking exer- until age 60, and (b) the rate of stunting for children cise can spur policy action is strongly influenced by under age 5. Adult survival rates can be interpreted the over-time and cross-country coverage of the as a proxy for the range of fatal and nonfatal health metric. Aiming for good coverage limits the com- outcomes that a child born today would experi- ponents of the index to data that are systematically ence as an adult if current conditions prevail into available for a large number of countries over time. the future. Stunting is broadly accepted as a proxy Yet, for an index to promote change, the compo- for the prenatal, infant, and early childhood health nents of the HCI should be responsive to policy environment, and so summarizes the risks to good 2 As a result of the criteria for its construction, the index measures dimensions of human capital that are important, but not all of the important dimensions of human capital are included in the index. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 3 Box 1.1: Learning-Adjusted Years of Schooling The knowledge and skills that an individual acquires through schooling form an important part of her human capital. However, the standard summary measure for education used in aggre- gate-level contexts—the average number of years of schooling in a population—is an imprecise proxy for education, because a given number of years in school leads to much more learning in some settings than in others. As recent research shows, students in different countries who have completed the same number of years of school often have vastly different learning outcomes.1 The Learning-Adjusted Years of Schooling (LAYS), a measure described in Filmer et al. (2018), addresses this concern by combining information on the quantity and quality of schooling into a single easy-to-understand metric of progress. It is calculated as the product of average years of school and a particular measure of learning relative to a numeraire: ​LAYS​C​= ​SC​ ​× ​Rc​n​​​   (1) where ​SC​ ​is a measure of the average years of schooling acquired by a relevant cohort of the pop- ulation of country ​c​, and ​Rc​n​​ is a measure of learning for a relevant cohort of students in country​ c​, relative to a numeraire (or benchmark). For the HCI, Expected Years of School measures the quantity of education. Harmonized Test Scores from the 2020 update of the Global Dataset on Education Quality provides information on education quality relative to a benchmark score of 625, which corresponds to the Trends in International Mathematics and Science Study (TIMSS) standard of advanced achievement: ​HTS​c​ ​LAYS​C​= ​EYS​C​× ​_ 625 ​​   (2) By adjusting years of school for quality, LAYS reflects the reality that children in some coun- tries learn far less than those in other countries, despite being in school for a similar amount of time. The simplicity and transparency of its construction make LAYS a compelling sum- mary measure of education to use in policy dialogue.2 Filmer et al. (2018) also find that LAYS improves upon the standard metric of average years of schooling as a predictor of economic growth. Source: Filmer et al. (2018). a In Nigeria, for example, 19 percent of young adults who have completed only primary education are able to read; by contrast, 80 percent of Tanzanians in the same category are literate (Kaffenberger and Pritchett 2017, as reproduced in World Bank 2018). b Like all aggregate measures, LAYS should be used with caution. Because there are standard errors around test measures, any LAYS measure will also have some error band around it. This means that it is important not to overinterpret small cross-country differences or small changes over time. health that children born today are likely to expe- makes it challenging to combine the different rience in their early years—with important conse- components into a single index. Rather than rely- quences for health and well-being in adulthood. ing on ad hoc aggregation with arbitrary weights, the HCI uses the estimated earnings associated The health and education components of human with an additional unit of health and education to capital have intrinsic value that is undeniably translate them into contributions to worker pro- important but difficult to quantify. This in turn ductivity, relative to a benchmark of complete 4 The Hum an Capita l Index 20 20 U pdate Table 1.1: Human Capital Index 2020, averages by World Bank region Middle Europe & Latin East & Sub- East Asia Central America & North North South Saharan Indicator & Pacific Asia Caribbean Africa America Asia Africa HCI Component 1: Survival Probability of Survival to Age 5 0.98 0.99 0.98 0.98 0.99 0.96 0.93 HCI Component 2: School Expected Years of School 11.9 13.1 12.1 11.6 13.3 10.8 8.3 Harmonized Test Scores 432 479 405 407 523 374 374 HCI Component 3: Health Survival Rate from Age 15 to 60 0.86 0.90 0.86 0.91 0.91 0.84 0.74 Fraction of Children Under 5 Not 0.76 0.90 0.85 0.82 – 0.69 0.69 Stunted Human Capital Index (HCI) 2020 0.59 0.69 0.56 0.57 0.75 0.48 0.40 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The table reports averages of the index components and the overall HCI by World Bank Group regions. “—” indicates data are unavailable. education and full health (see Box 1.2).3 The result- 1.2 THE HUMAN CAPITAL INDEX 2020 ing index ranges between 0 and 1. A country in which a child born today can expect to achieve full The HCI 2020 is reported later in Table 1.2. health (no stunting and 100 percent adult survival) Country scores are sorted from lowest to highest. and full education potential (14 years of high-qual- Next to the HCI score, lower and upper bounds for ity school by age 18) would score a value of 1. the estimates are reported. Unlike the HCI 2018 Therefore, a score of 0.70 indicates that the pro- launch, countries’ rankings are not reported, for ductivity as a future worker of a child born today is reasons that are detailed in Box 1.6. 30 percent below what could have been achieved with complete education and full health. Because The sobering reality is that, as measured by the the theoretical underpinnings of the HCI are in Human Capital Index 2020, worldwide, a child the development accounting literature, the index born today would expect to achieve on average is linked to real differences in how much income only 56 percent of her full productivity as a future a country can generate in the long run (see Box 1.3 worker. This is before accounting for any impact for limitations of the HCI). If a country has a score that may have resulted from the COVID-19 pan- of 0.50, then the gross domestic product (GDP) demic. Clearly there is considerable heterogeneity per worker could be twice as high if the country around the 56 percent figure. Importantly, the HCI reached the benchmark of complete education is lower in poor countries than in rich countries and full health (see appendix A for a detailed dis- by a substantial margin. In the poorest countries cussion of the HCI methodology). in the world, a child born today will grow up to 3 The literature has recognized the usefulness of moving from “a large and eclectic dashboard” to a single summary metric (Sti- glitz, Sen, and Fitoussi 2009). However, doing so requires a coherent aggregation method, in contrast with “mashup indicators of development” that combine different components in arbitrary ways (Ravallion 2010). The HCI is constructed by transforming its components into contributions to productivity, anchored in microeconometric evidence on the effects of education and health on worker productivity, consistent with the large literature on development accounting (see, for example, Caselli 2005). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 5 Box 1.2: The HCI’s aggregation methodology The components of the HCI are combined into a single index by first converting them into contri- butions to productivity relative to a benchmark of complete education and full health. Multiplying these contributions to productivity together gives the overall HCI: HCI = Survival × School × Health    (1) In the case of survival, the relative productivity interpretation is stark: children who do not survive childhood never become productive adults. As a result, expected productivity as a future worker of a child born today is reduced by a factor equal to the survival rate, relative to the benchmark where all children survive. ______________ 1 − Under-5 Mortality Rate ​Survival =    ​ 1 ​   (2) The benchmark of complete high-quality education corresponds to 14 years of school and a har- monized test score of 625. The relative productivity interpretation for education is anchored in the large empirical literature measuring the returns to education at the individual level. A rough consensus from this literature is that an additional year of school raises earnings by about 8 per- cent. The parameter ​ϕ​= 0.08 measures the returns to an additional year of school and is used to convert differences in learning-adjusted years of school across countries into differences in worker productivity. Harmonized Test Score ​School = ​e​ϕ​(​Expected Years of School × ​       ​−14​)​ ___________ 625 ​   (3) Compared with a benchmark where all children obtain a full 14 years of school by age 18, a child who obtains only 10 years of education can expect to be 32 percent less productive as an adult (a gap of 4 years of education, multiplied by 8 percent per year). In the case of health, the relative productivity interpretation is based on the empirical literature measuring the economic returns to better health at the individual level. The key challenge in this literature is that there is no unique, directly measured summary indicator of the various aspects of health that matter for productivity. This microeconometric literature often uses proxy indicators for health, such as adult height. This is because adult height can be measured directly and reflects the accumulation of shocks to health through childhood and adolescence. A rough consensus drawn from this literature is that an improvement in health associated with a one-cen- timeter increase in adult height raises productivity by 3.4 percent. Converting this evidence on the returns to one proxy for health (adult height) into the other proxies for health used in the HCI (stunting and adult survival) requires information on the rela- tionships between these different proxies: • For stunting, there is a direct relationship between stunting in childhood and future adult height, because growth deficits in childhood persist to a large extent into adulthood, together with the associated health and cognitive deficits. Available evidence suggests that a reduc- tion in stunting rates of 10 percentage points increases attained adult height by approxi- mately one centimeter (0.1 × 10.2), which increases productivity by 3.5 percent. • For adult survival, the empirical evidence suggests that, if overall health improves, both adult height and adult survival rates increase in such a way that adult height rises by 1.9 centi- meters for every 10-percentage-point improvement in adult survival. This implies that an 6 The Hum an Capita l Index 20 20 U pdate improvement in health that leads to an increase in adult survival rates of 10 percentage points is associated with an improvement in worker productivity of 1.9 × 3.4 percent, or 6.5 percent. In the HCI, the estimated contributions of health to worker productivity based on these two alternative proxies are averaged together, if both are available, and are used individually if only one of the two is available. The contribution of health to productivity is expressed relative to the benchmark of full health, defined as the absence of stunting, and a 100 percent adult survival rate. For example, compared with a benchmark of no stunting, in a country where the stunting rate is 30 percent, poor health reduces worker productivity by 30 x 0.34 percent or 10 percent. ​Health = ​e​​(​γASR ​ ​×​(Adult Survival Rate−1)​+​γStunting ​ ​×(​ Not Stunted Rate−1)​)​/2 ​​   (4) Compared with the benchmark of 100 percent adult survival, poor health reduces worker pro- ductivity by (30 x 0.65) percent, or 19.5 percent, in a country where the adult survival rate is 70 percent. The average of the two estimates of the effect of health on productivity is used in the HCI. These parameters used to convert the components of the index into their contributions to pro- ductivity (​ϕ​= 0.08 for school, γ​ ASR ​ ​​= 0.65 for adult survival, and γ​ Stunting ​ ​= 0.35 for stunting) serve as weights in the construction of the HCI. The weights are chosen to be the same across countries, so that cross-country differences in the HCI reflect only cross-country differences in the com- ponent variables. This facilitates the interpretation of the index. This is also a pragmatic choice, because estimating country-specific returns to education and health for all countries included in the HCI is not feasible. Source: Kraay (2018). be only 30 percent as productive as she could be, expect to be only 58 percent as productive as a while in the richest countries the corresponding future worker as a child in Europe and Central figure is 80 percent or more (see Figure 1.1, which Asia (see Table 1.1). plots the HCI 2020 on the vertical axis against log GDP per capita at PPP on the horizontal axis). The correlation between poverty and low HCI scores is also high. Given that better education While the correlation between the HCI and GDP and health translate to improved productivity for per capita is high, some economies perform sig- people, and that human capital is often the only nificantly better than their income levels might asset the poor have, the World Bank’s twin goals suggest. These include Estonia, Kyrgyz Republic, of shared prosperity and eradicating extreme pov- Vietnam, and West Bank and Gaza. Conversely, erty are unlikely to be met without human capital in a number of countries, human capital is lower improvements. Accordingly, the world’s extreme than per capita income would suggest. Among poor are disproportionately found in countries these are a few resource-rich countries, where with the lowest HCI; 30 percent of the world’s poor human capital has not yet matched the potential reside in the 10 countries with the lowest HCI val- that one would envisage given these countries’ ues, although these 10 countries are home to only 5 development. A child in Sub-Saharan Africa can percent of the total global population (Figure 1.2). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 7 Figure 1.1: The Human Capital Index 2020 1.0 Japan 0.8 Korea, Rep. Estonia HCI, circa 2020 Belarus Vietnam Luxembourg Qatar Uzbekistan 0.6 Kyrgyz Republic West Bank and Gaza Saudi Arabia Kuwait Panama Botswana Iraq 0.4 Eswatini 0.2 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data and the World Development Indicators and Penn World Tables 9.1 for per capita GDP data. Notes: The figure uses real GDP per capita at PPP, in constant 2011 US$, for most recently available data as of 2019. Per capita GDP data for South Sudan are not available. The figure plots the country-level HCI on the y-axis and GDP per capita in PPP on the x-axis. The dashed line illustrates the fitted regression line between GDP per capita and the HCI 2020. Scatter points above (below) the fitted regression line illustrate economies that perform higher (lower) in the HCI than their level of GDP would predict. Countries above the 95th and below the 5th percentile in distance to the regression fitted line are labeled. Figure 1.2: Concentration of the extreme poor in economies sorted by their Human Capital Index 1.0 Cumulative share of the world’s poor 0.8 80% of the world’s poor reside in economies with an HCI under 0.5 (US$1.9 2011 PPP) 0.6 0.4 30% of the world’s poor reside in the bottom 10 economies sorted by HCI 0.2 0 0 .2 .4 .6 .8 1 Share of world population, economies sorted by their HCI 2020 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Poverty values come from Corral et al. (2020) and are calculated pre-COVID-19. Notes: The figure corresponds to the 174 economies for which a Human Capital Index has been calculated. It covers roughly 98 percent of the world’s population. Economies are sorted by their HCI value. The horizontal axis represents the share of the global population accounted for by the countries once sorted. 8 The Hum an Capita l Index 20 20 U pdate Figure 1.3: Human Capital Index 2020 components, distribution by country income group a. Probability of Survival to Age 5 b. Expected Years of School 1.0 14 Probability of Survival to Age 5, circa 2020 Seychelles Panama Nauru 12 Expected Years of School, Mauritius Nauru Palau Palau 0.95 Trinidad Panama and circa 2020 10 Tobago Gabon 8 Botswana 0.9 Iraq Nigeria 6 0.85 4 c. Harmonized Test Scores d. Fraction of Children Under 5 Not Stunted 600 1.0 Fraction of Children Under 5 Not Stunted, Harmonized Test Scores, circa 2020 Vietnam 500 0.8 circa 2020 Marshall Islands 400 0.6 Guatemala Nauru 300 0.4 e. Adult Survival Rate 1.0 Low income Lower-middle income Adult Survival Rate, circa 2020 0.9 Upper-middle income High income 0.8 Namibia 0.7 Marshall Islands South Africa Sierra Leone 0.6 Eswatini Central African Republic Lesotho 0.5 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: Each box spans the interquartile range with the upper and lower end of the boxes illustrating the 25th and 75th percentile values. The horizontal lines in the inner boxes represent the median value. Outer horizontal lines show maximum and minimum values excluding outliers. Thinner box plots indicate less dispersion in values. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 9 Figure 1.4: Decomposition of observed mean HCI differences between selected country income groups Upper middle vs High Lower middle vs High Low vs High 0 0.1 0.2 0.3 0.4 HCI points contributed to observed difference Child survival Harmonized test scores Expected years of school Health Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the contribution to observed HCI differences between income groups. In fact, 80 percent of the world’s extreme poor points, almost 25 are accounted for by the differ- reside in countries with an HCI under 0.5. If pros- ences in expected years of school and harmonized perity is to be shared, growth must be inclusive test scores. Overall, differences in the quality for those at the bottom of the distribution, and and quantity of schooling account for the largest inclusive growth necessitates strong investments in share of index differences across country income human capital. groups, ranging from 65 to 85 percent. Two elements help explain how different dimen- There is also considerable heterogeneity within sions of human capital contribute to overall country income groups, and the difference in HCI cross-country differences in the HCI. The first are between the country with the lowest and the coun- the weights of the health and education compo- try with the highest HCI in each income group nents of the HCI, reflecting the empirical litera- rivals the differences between income groups and, ture on the contribution of health and education in some cases, exceeds it. For example, the differ- to earnings (Box 1.2 and appendix A). Second, the ence in the HCI between the top and bottom per- components have different distributions, globally formers among high-income economies is roughly and by country income groupings, according to the 0.38, or 38 HCI points. This compares with a differ- World Bank most recent classification. For exam- ence of 33 points between the average HCI values ple, the variation of child survival is nine times of high- and low-income countries. Overall, both larger among low-income than among high-in- within and across all groups, education still accounts come countries, where child survival is uniformly for the largest share of the differences observed close to 100 percent (Figure 1.3). between top and bottom performers (Figure 1.5). However, education accounts for a smaller share A simple decomposition exercise can help account as one moves down income groups, falling from for differences in the HCI across country income roughly 90 percent among high-income to 60 per- groups. Consider the HCI difference between the 4 cent among low-income economies. In contrast, typical low-income and high-income country, differences in child survival rates account for less of which is about 0.33 (Figure 1.4). Of these 33 HCI the difference in HCI scores among high-income 4 The decomposition of the group averages is obtained via a Shapley decomposition. For an application see Azevedo, Inchauste, and Sanfelice 2013. 10 The Hum an Capita l Index 20 20 U pdate Figure 1.5: Differences between the top and bottom HCI performers within each country income group High-income economies Upper-middle-income economies Lower-middle-income economies Low-income economies 0 0.2 0.4 0.6 0.8 1.0 Share of difference between top and bottom performer explained by each component Child survival Harmonized test scores Expected years of school Health Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the share of the observed HCI differences between selected economies by component. Comparison economies for the high-income group are Singapore and Panama; for the upper-middle-income group Belarus and Iraq; for lower-middle-income economies Nigeria and Vietnam; and for low-income economies Central African Republic and Tajikistan. countries, largely because countries in this group differences across socioeconomic quintiles within are close to universal child survival. The same is true countries account for nearly one-third of the total for the health component, with stunting and adult variation in human capital.7 Outcomes can also survival taken together for easy comparison. Health vary across rural-urban status, as in the case of differences explain a lower share of HCI differences Romania. In some of the country’s counties, there as one moves from low- to higher-income coun- are urban areas with learning outcomes as high as tries, since health outcomes tend to be uniformly top performers in Europe, while some rural areas better as countries get richer. These results reflect 5 rank at par with countries in the bottom third of the the fact that, within high-income groups, values for HCI distribution.8 Some of these within-country health and survival components in most countries differences aIign with ethnic divides. For example, are close to the frontier, whereas there is still con- in Vietnam, survey data from 2014 disaggregated siderable variation in test scores across countries. 6 by ethnic group show that ethnic minorities have an HCI score of 0.62, compared with 0.75 for the Gaps in human capital outcomes between rich and ethnic-majority Kinh. At 32 percent, stunting rates poor people within countries can be quite large. are two times larger among ethnic minorities than A socio-economic disaggregation of the HCI, among the Kinh majority. School enrollment also constructed using comparable survey data for 50 lags among ethnic minorities relative to their Kinh low- and middle-income countries, revealed that peers by 30 percentage points.9 5 Among upper-middle-income economies, the health component value of the bottom performer is higher than that of the top performer, and thus it accounts for a negative share of the difference. 6 See the box plots in Figure 1.3. 7 World Bank (2019b). 8 Lucchetti, Badiani-Magnusson, and Ianovici (2019). 9 World Bank (2019b). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 11 Box 1.3: Limitations of the HCI Like all cross-country benchmarking exercises, the HCI has limitations. Components of the HCI such as stunting and test scores are measured only infrequently in some countries and not at all in others. Data on test scores come from different international testing programs that need to be converted into common units, and the age of test-takers and the subjects covered vary across testing programs. Moreover, test scores may not accurately reflect the quality of the whole edu- cation system in a country, to the extent that test-takers are not representative of the popula- tion of all students. Reliable measures of the quality of tertiary education that are comparable across most countries of the world do not yet exist, despite the importance of higher education for human capital in a rapidly changing world. The data on enrollment rates needed to esti- mate expected years of school often have many gaps and are reported with significant lags. Socioemotional skills are not explicitly captured. Child and adult survival rates are imprecisely estimated in countries where vital registries are incomplete or nonexistent. These limitations have implications not only the construction of the 2020 update but also the comparison of the index over time. One objective of the HCI is to call attention to these data shortcomings and to galvanize action to remedy them. Improving data will take time. In the interim and in recognition of these limitations, the HCI should be interpreted with caution. The HCI provides rough estimates of how current education and health will shape the productivity of future workers but is not a finely graduated measurement that can distinguish small differences between countries. Naturally, because the HCI captures outcomes, it is not a checklist of policy actions, and the proper type and scale of interventions to build human capital will be different in different countries. Although the HCI combines education and health into a single measure, it is too blunt a tool to inform the cost-ef- fectiveness of policy interventions in these areas, which should instead be assessed based on careful cost-benefit analysis and impact assessments of specific programs. Because the HCI uses common estimates of the economic returns to health and education for all countries, it does not capture cross-country differences in how well countries are able to productively deploy the human capital they have. Finally, the HCI is not a measure of welfare, nor is it a summary of the intrinsic values of health and education; rather, it is simply a measure of the contribution of current health and education outcomes to the productivity of future workers. 1.3 HCI 2020 UPDATE—INDEX COMPONENTS Child survival The probability of survival to age 5 is calculated 1.3.1 HCI components and data sources as the complement of the under-5 mortality rate. The components of the Human Capital Index are The under-5 mortality rate is the probability of a built using publicly available official data, prima­ child born in a specified year dying before reach- rily from administrative sources. The data are sub- ing the age of 5 if subject to current age-specific ject to a careful vetting process with World Bank mortality rates. It is frequently expressed as a country teams and, at the discretion of country rate per 1,000 live births, in which case it must teams, with line ministry counterparts. These data be divided by 1,000 to obtain the probability of and the relevant definitions are described below dying before age 5. Under-5 mortality rates are and in more in detail in appendix C. calculated by the United Nations Interagency 12 The Hum an Capita l Index 20 20 U pdate Group for Child Mortality Estimation (IGME) Harmonized test scores based on mortality as recorded in household sur- The school quality indicator is based on a large- veys and vital registries. For the 2020 update of scale effort to harmonize international student the HCI, under-5 mortality rates come from the achievement tests from several multicountry September 2019 update of the IGME estimates testing programs to produce the Global Dataset and are available at the Child Mortality Estimates on Education Quality. A detailed description of website.10 the test score harmonization exercise is provided in Patrinos and Angrist (2018), and the HCI draws Expected years of school on an updated version of this dataset as of January The expected years of school (EYS) component 2020. The dataset harmonizes scores from three of the HCI captures the number of years of major international testing programs: the Trends school a child born today can expect to obtain in International Mathematics and Science Study by age 18, given the prevailing pattern of enroll- (TIMSS), the Progress in International Reading ment rates in her country. Conceptually, EYS is Literacy Study (PIRLS), and the Programme for the sum of enrollment rates by age from ages 4 International Student Assessment (PISA). It further to 17. Because age-specific enrollment rates are includes four major regional testing programs: neither broadly nor systematically available, data the Southern and Eastern Africa Consortium for on enrollment rates by level of school are used Monitoring Educational Quality (SACMEQ), the to approximate enrollment rates in different age Program for the Analysis of Education Systems brackets. Pre-primary enrollment rates approxi- (PASEC), the Latin American Laboratory for mate the enrollment rates for 4- and 5-year-olds, Assessment of the Quality of Education (LLECE), primary enrollment rates approximate the rates and the Pacific Island Learning and Numeracy for 6- to 11-year-olds, lower-secondary rates Assessment (PILNA). It also incorporates Early approximate for 12- to 14-year-olds, and upper Grade Reading Assessments (EGRAs) coordinated secondary rates approximate for 15- to 17-year- by the United States Agency for International olds. Cross-country definitions in school start- Development. The 2020 update of the Global ing ages and the duration of the various levels Dataset on Education Quality extends the data- of school imply that these will only be approxi- base to 184 countries and economies from 2000 to mations of the number of years of school a child 2019, drawing on a large-scale effort by the World can expect to complete by age 18. Enrollment Bank to collect learning data globally. Updates rates for 2020 for each school level and for dif- to the database come from new data from PISA ferent enrollment rate types are obtained from 2018, PISA for Development (PISA-D),13 PILNA, the UNESCO Institute for Statistics (UIS). 11 UIS and EGRA. The database adds 20 new countries.14 data were then complemented with inputs from This brings the percentage of the global school- World Bank teams working on specific countries age population represented by the database to validate the data and provide more recent val- to 98.7 percent. In addition, more recent data ues when available.12 points have been added for 94 countries (Angrist, 10 http://www.childmortality.org/. 11 http://data.uis.unesco.org/. See Appendix C for the description of different enrollment rates: gross, net, adjusted net, and total net enrollment rates. 12 For the 2020 update, this review process was conducted between January and May 2020 in collaboration with the country units of the World Bank. 13 PISA-D results are only used for Panama and Bhutan. 14 For the 20 new countries included in the Global Dataset on Education Quality, eight are updated using EGRAs, eight using PILNA, three using PISA and PISA-D, and one using a national TIMSS-equivalent assessment. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 13 Iqbal, and Patrinos, 2020).15 Since the launch of Adult survival rates the HCI in 2018, a complementary measure has The adult survival rate is calculated as the com- been created to address foundational skills and plement of the mortality rate for 15- to 60-year- to help countries prioritize their response to HCI olds. The mortality rate for 15- to 60-year-olds and LAYS scores: Learning Poverty, the share of is the probability of a 15-year-old in a specified 10-year-olds who cannot read and understand year dying before reaching the age of 60 if sub- a simple story (see Box 1.4). The correlation ject to current age-specific mortality rates. It is fre- between Learning Poverty and LAYS is very high, quently expressed as a rate per 1,000 alive at 15, in in the range of −0.90. The Learning Poverty mea- which case it must be divided by 1,000 to obtain sure is available for 113 of the economies with an the probability of a 15-year-old dying before age HCI 2020. 60. Adult mortality rates for the 2020 update of the HCI come from the 2019 update of the UNPD Fraction of children under 5 not stunted World Population Prospects estimates, available The fraction of children under 5 not stunted is cal- at the World Population Prospects website.18 Since culated as the complement of the under-5 stunt- UNPD does not individually report adult mor- ing rate. The stunting rate is defined as the share of tality rates for countries with less than 90,000 children under the age of 5 whose height is more inhabitants, data from the UNPD are supple- than two reference standard deviations below the mented with adult mortality rates from the Global reference median for their ages. The reference Burden of Disease (GBD) project, managed by the median and standard deviations are set by the Institute of Health Metrics and Evaluation (IHME). World Health Organization (WHO) for normal Data from this source are used for Dominica and healthy child development. 16 Child-level stunting The Republic of the Marshall Islands. Data for prevalence is averaged across the relevant 0–5 Nauru, Palau, San Marino, St. Kitts and Nevis, and age range to arrive at an overall under-5 stunting Tuvalu come from the World Health Organization rate. The stunting rate is used as a proxy for latent (WHO). The GBD data for the HCI 2020 come health of the population, in addition to the adult from the GBD 2017 update and can be retrieved survival rate, in countries where stunting data are from the IHME data visualization site.19 The WHO available, as discussed below. Stunting rates for data are located on the UN Data platform.20 the 2020 update of the HCI come from the March 2020 update of the Joint Malnutrition Estimates 1.3.2 Index components across countries ( JME) database, available at the UNICEF website.17 All five components of the index increase with This latest update to the database allows an update income, though at a different pace (Figure 1.6). of stunting rates for 54 countries, and adds stunt- Child survival rates range from 0.998 (2 deaths per ing rates for Argentina, Bulgaria, and Uzbekistan, 1,000 live births) in the richest countries to around which did not have a rate in the previous iteration 0.880 (120 deaths per 1,000 live births) in the poor- of the HCI. est countries, reflecting the disproportionate bur- den of child mortality that low-income countries 15 Of the 94 countries with updated test scores in the Global Dataset on Education Quality, seventy-five are from PISA 2018, seven from PISA-D, five from EGRAs, and seven from PILNA. 16 World Health Organization (2009). 17 UNICEF/WHO/The World Bank Group Joint Child Malnutrition Estimates: Levels and Trends in Child Malnutrition: Key Findings of the 2020 Edition. https://www.who.int/publications-detail/jme-2020-edition. 18 United Nations Population Division, https://population.un.org/wpp/. 19 http://www.healthdata.org/results/data-visualizations. 20 https://data.un.org/. 14 The Hum an Capita l Index 20 20 U pdate Box 1.4: Measuring Learning Poverty The World Bank collaborated with the UNESCO Institute for Statistics (UIS) to create a measure of Learning Poverty—the share of 10-year-olds who cannot read and understand a simple story. Using a database developed jointly with UIS, the Bank estimates that 53 percent of children in low- and middle-income countries suffer from learning poverty. In the poorest countries, the number is often more than 80 percent. Such high levels of learning poverty are an early warning sign that the LAYS indicator, which measures quantity and quality of education that 18-year-olds have benefited from, will be unacceptably low for that cohort of children in a few years. In high- er-performing systems, virtually all children learn to read with comprehension by age 10. While it may take decades to build up the high-quality education systems that lead to the highest scores on the LAYS indicator of the HCI, teaching children to reach a minimum proficiency in reading requires much less time. Why reading? Children need to learn to read so that they can read to learn. Those who do not become proficient in reading by the end of primary school often cannot catch up later, because the curriculum of every school system assumes that secondary-school students can learn through reading. Reading is, in other words, a gateway to all types of academic learning. This is not to say that reading is the only skill that matters. Reading proficiency can serve as a proxy or warning indicator for foundational learning in other areas that are also essential, like math- ematics and reasoning abilities. Education systems that enable all children to read are likely to succeed in helping them learn other subjects as well. Across countries and schools, the data show that proficiency rates in reading are highly correlated with proficiency in other subjects. How is learning poverty calculated? Conceptually similar to the LAYS indicator in the HCI for youth, the Learning Poverty measure combines learning with enrollment, to emphasize the impor- tance of learning for all children and not just those currently in school. The learning component captures enrolled students who cannot read with comprehension, while the participation com- ponent corresponds to the out-of-school rate. “Reading with comprehension” is defined here as reaching the global minimum proficiency in literacy. The UIS leads the Global Alliance to Monitor Learning (GAML), which agreed to a common definition of minimum proficiency in literacy for the purposes of monitoring Sustainable Development Goal (SDG) 4. With this definition, several cross-national and some national assessments were harmonized by applying GAML’s definition of reading proficiency as a common benchmark. Unlike the HCI, Learning Poverty relies only on assessments targeting children from grades 4 to 6. For each assessment incorporated into the database, the harmonization process looks at the definitions of each level of proficiency for that exam and selects the one that maps most clearly to the GAML definition. The harmonization process allowed much greater coverage of countries than relying on a single assessment like the Progress in International Reading Literacy Study (PIRLS)—an excellent assessment for mea- suring Learning Poverty, but one in which relatively few low- and middle-income countries partic- ipate. The high correlation between students’ performance on different assessments increased confidence that this harmonization method is valid. Once the share of children below minimum proficiency is calculated, the final step in calculating Learning Poverty is to adjust this share for out-of-school children of primary-school age who are considered nonproficient in reading. The HCI, LAYS, and Learning Poverty, each with its own unique mandate and methodology, are synthetic indicators intended to build political commitment and galvanize action. Sources: World Bank Education Global Practice and World Bank (2019a). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 15 Figure 1.6: Human Capital Index 2020—index components 1.0 14 Probability of Survival to Age 5, circa 2020 Nicaragua Kyrgyz Republic Kyrgyz Republic Expected Years of School, circa 2020 Solomon Islands Tuvalu Vanuatu Nepal 12 Micronesia, Fed. Sts. Rwanda Haiti Kiribati Malawi Zimbabwe 0.95 Burundi 10 Malawi Mauritania Congo, Dem. Rep. Angola Côte d’Ivoire Angola Gabon Benin 8 Botswana Mali 0.9 Guinea Iraq Sudan Eswatini Chad Nigeria 6 Mali 0.85 4 Liberia 6 8 10 12 6 8 10 12 Log GDP per capita at PPP, circa 2020 Log GDP per capita at PPP, circa 2020 Fraction of Children Under 5 Not Stunted, circa 2020 600 1.0 St. Lucia North Macedonia Turkey Samoa Bulgaria Harmonized Test Scores, circa 2020 West Bank and Gaza Kazakhstan Argentina Estonia Poland Vietnam Gambia, The Brunei Darussalam 500 0.8 Haiti Malaysia Uzbekistan Ukraine Kenya Cambodia Qatar Burundi Saudi Arabia Angola 400 0.6 Kuwait Timor−Leste Panama Guatemala Dominican Republic Papua New Guinea South Africa Nigeria Ghana 300 0.4 6 8 10 12 6 8 10 12 Log GDP per capita at PPP, circa 2020 Log GDP per capita at PPP, circa 2020 1.0 Morocco Nauru West Bank and Gaza 0.9 Adult Survival Rate, circa 2020 Timor−Leste Vanuatu Tajikistan Solomon Islands Nepal 0.8 Namibia 0.7 South Africa Nigeria Côte d’Ivoire Zimbabwe Eswatini 0.6 Central African Republic Lesotho 0.5 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure reports the most recent cross-section of 174 economies for the five HCI components (child survival, expected years of school, harmonized test scores, fraction of children under 5 not stunted, and adult survival), as used to calculate the 2020 HCI. Each panel plots the country-level averages for each component on the y-axis and GDP per capita in PPP on the x-axis. The dashed line illustrates the fitted regression line between GDP per capita and the respective component. Scatter points above (below) the fitted regression line illustrate economies that perform higher (lower) in the outcome variable than their level of GDP would predict. Countries above the 95th and below the 5th percentile in distance to the fitted regression line are labeled. 16 The Hum an Capita l Index 20 20 U pdate continue to face. Child survival rates also vary sig- top of the distribution and low-income countries nificantly by region, with economies in the Europe are at the bottom of the distribution. However, and Central Asia region bundled at the top of the in economies like Malawi, Zimbabwe, Nepal, or distribution and the lowest rates in Sub-Saharan the Kyrgyz Republic, expected years of school Africa, in countries like Chad, Nigeria, and Sierra are higher than their level of GDP would predict, Leone. However, in a number of economies in reflecting the progress these countries were able to Sub-Saharan Africa, including Burundi, Malawi, or make in improving access to schooling (Figure .16). Rwanda, child survival rates are significantly higher than their level of GDP would predict (Figure 1.6). Outliers where the quantity of schooling is about 2.5 to 5.3 years below what their level of GDP would While internationally comparable stunting mea- predict include economies such as Iraq, Mali, and sures are primarily collected in low- and mid- Liberia, which are characterized by different lev- dle-income countries, the share of stunted chil- els of institutional fragility and conflict. Quality dren decreases as countries get richer. However, of schooling—as measured by harmonized test income and stunting rates do not always go in scores (HTS)—increases with income, too, though lockstep, including across socioeconomic groups seemingly faster than years of education. The HTS within countries.21 For example, in countries such ranges from a score of around 305 in the poorest as Burundi, Niger, and Tanzania, the gap in stunt- countries to a score of around 575 in the richest ing rates between the 1st and the 4th socioeconomic countries (Figure 1.6). To interpret the units of the quintiles is smaller than the gap between stunting HTS, note that 400 corresponds to the benchmark rates in the 4th and 5th quintiles (the richest house- of “low proficiency” in TIMSS at the student level, holds), reflecting the interaction of environmental, while 625 corresponds to “advanced proficiency.” economic, and cultural factors that can contribute Accounting for the level of GDP, economies such to slower physical development in children. 22 In as Vietnam, Ukraine, and Uzbekistan, as well as countries such as Papua New Guinea, Timor-Leste, Kenya and Cambodia, performed particularly well and Guatemala, more than 45 percent of children in learning. Vietnam reaches an HTS of 519, a level are stunted. On the other end of the spectrum are similar to countries like Sweden, the Netherlands, economies like Samoa, Tonga, Moldova, or West and New Zealand, which are significantly richer.23 Bank and Gaza, where the stunting rate is below Economies where learning is below what their 10 percent, and significantly lower than their income per capita would predict include high-in- level of GDP would predict. The second proxy for come countries such as Kuwait, Saudi Arabia, and health—adult survival—is lowest in Lesotho, Qatar. Their relatively disappointing performance Eswatini, and the Central African Republic, where in learning may result in part from a traditional the chances of surviving from age 15 to age 60 are emphasis on investing in school infrastructure at 60 percent or lower. rather than other factors that are also necessary to improve educational outcomes. These include Quantity of schooling—as measured by Expected governance and accountability, effective mon- Years of School (EYS)—increases as countries get itoring mechanisms, information sharing with richer. High-income countries are bundled at the parents and students, and school systems geared 21 de Onis and Branca (2016). 22 World Bank (2019b). 23 Note that Vietnam enters the HCI 2020 with its 2015 PISA score, since 2018 PISA scores are not reported for the country. While Vietnam participated in the 2018 round of PISA using paper-based instruments, the OECD’s country note states that the interna- tional comparability of the country’s performance in reading, mathematics, and science could not be fully ensured (OECD 2019). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 17 Figure 1.7: Sex-disaggregated Human Capital Index and its components Human Capital Index Probability of Survival to Age 5 Expected Years of School Harmonized Test Scores Fraction of Children Under 5 Not Stunted Adult Survival Rate 0 .2 .4 .6 .8 1 Girls to boys ratio, circa 2020 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The red vertical line indicates gender parity for each component. Simple averages are computed without population weights. toward inclusive learning.24 Education systems in compared to full potential far exceeds any gender these countries may also be reacting to the pull gap in HCI in most countries. Boys and girls are from labor markets, where pervasive informality 2.6 and 2.5 years of schooling away from com- generates low returns to schooling, and the lure pleting upper-secondary education. A large share of public employment puts more emphasis on of boys and girls are stunted, 24 and 21 percent, diplomas than on skills.25 As a consequence, learn- respectively. Far too many boys and girls do not ing lags behind the progress that countries in this survive beyond their fifth birthday, 2.8 and 2.4 region have achieved in access to schooling and percent, respectively. Conditional on making it to gender parity. age 15, only 83 percent of boys and 89 percent of girls are expected to survive to age 60. 1.4 HCI MEASURES OF GENDER GAPS IN The global HCI average, however, masks import- HUMAN CAPITAL ant regional and income-group differences with respect to gender (Figure 1.8).. While girls still Globally, the average HCI for girls is slightly higher surpass boys in the HCI value overall, with lower (0.59) than that for boys (0.56).26 This pattern can stunting as well as lower child and adult mortality be observed across all HCI components (Figure 1.7). rates in all regions and income groups, advantages for girls are more prominent in some regions and While the gap between boys and girls has closed muted in others. For example, the gap in stunting in these early-life outcomes, boys and girls both rates between girls and boys is as high as 4.6 per- remain far from the frontier of complete edu- centage points in Sub-Saharan Africa, with boys cation and full health. The gap in human capital having a higher stunting rate. 24 Galal et al. (2008). 25 World Bank (2013) and El-Kogali and Krafft (2020). 26 This difference is statistically significant at the 5 percent level. 18 The Hum an Capita l Index 20 20 U pdate Figure 1.8: Regional and income-group variations in education gaps between boys and girls Low income Lower-middle income Upper-middle income High income 0 0.2 0.4 0.6 0.8 1 Girls to boys ratio, circa 2020 East Asia & Pacific Europe & Central Asia Latin America & Caribbean Middle East & North Africa North America South Asia Sub−Saharan Africa 0 0.2 0.4 0.6 0.8 1 Girls to boys ratio, circa 2020 Expected Years of School Harmonized Test Scores Source: World Bank calculations based on the 2020 update of the Human Capital Index. With regard to expected years of school, girls are The gender gap in the Human Capital Index var- still disadvantaged compared to boys in South ies quite widely across countries, with a difference Asia and Sub-Saharan Africa, where girls and in the score between boys and girls ranging from boys experience 0.45 and 0.15 years of school a low of -0.043 in Afghanistan to a high of 0.096 disadvantage, respectively (Figure 1.8). In set- in Lithuania (Figure 1.9). Overall, girls are outper- tings affected by fragility, conflict, and violence, forming boys in 140 of the 153 countries for which girls on average complete 0.14 years less school- sex-disaggregated data are available. ing than boys. In low-income countries, aside from completing less schooling, girls also have Expected years of school and harmonized test lower harmonized test scores, with a 0.8 percent scores show similar patterns. The gender gap in deficit. expected years of school favors boys in 46 countries T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 19 Figure 1.9: Country-level variations in gender gaps in HCI and education components Human Capital Index 0.1 Disparity favors girls 0.05 Girls−Boys Gap 0 Disparity favors boys −0.05 Expected Years of School Disparity favors girls 1 0 Girls−Boys Gap −1 −2 Disparity favors boys −3 Harmonized Test Scores 60 Disparity favors girls 40 Girls−Boys Gap 20 0 Disparity favors boys −20 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The X-axis show countries ranked by girl-boy gap in the variable in question. 20 The Hum an Capita l Index 20 20 U pdate (30 percent of all countries with a sex-disaggre- Overall, out of the 13 countries where boys have gated HCI, Figure 1.9). In learning outcomes, boys a higher HCI score than girls, eight are in Sub- are favored in 31 countries (20 percent). Although Saharan Africa, two in South Asia, one in the East expected years of school are higher for girls than Asia and Pacific region, one in Latin America and the for boys in most countries, in those countries where Caribbean, and one in the Middle East and North boys have an advantage over girls with respect to Africa region. Seven countries are low-income schooling, the magnitude of the resulting gen- countries, five are lower-middle-income countries, der disparity is larger. For example, in Tunisia, and one is an upper-middle-income country. In all Kiribati, and St. Vincent and the Grenadines, girls 13 countries, expected years of school for boys are on average complete more than one extra year higher than for girls, ranging from a quarter year in of school compared to boys, while in Angola and Peru to almost three full years in Afghanistan. On Afghanistan, boys on average complete 2.3 to average, boys have a 10-percentage-point higher 2.7 more years of school than girls. The top five likelihood of completing primary education, a countries where girls outperform boys in learning 12-percentage-point higher likelihood of com- outcomes are Nauru, Qatar, Oman, Bahrain, and pleting lower-secondary education, and a 13-per- Samoa, three of which are in the Middle East and centage-point higher likelihood of completing North Africa region. Conversely, 6 in 10 countries upper-secondary education. Boys also have bet- where boys have higher learning outcomes than ter learning outcomes than girls in nine of these girls are in Sub-Saharan Africa. In high- and mid- 13 countries. In Chad and Guinea, this difference dle-income countries, girls outperform boys in reaches more than 14 percent in favor of boys. enrollment and learning outcomes.27 For exam- ple, in Guyana, girls are expected to complete Human capital accumulation is a complex process. one-fifth of a year more schooling than boys with This is especially clear when looking at the HCI to 5 percent higher learning outcomes. This reverse understand gender gaps. Women, girls, men, and gap in enrollment begins in lower-secondary edu- boys face different challenges at different stages of cation and widens in upper-secondary, where girls the life cycle. The HCI focuses on specific life-cycle are 11 percent more likely to be enrolled than boys. stages in which girls have slight biological advan- tages over boys in child and adult survival rates.29 In survival and health outcomes, girls are generally As with any indicator, the components of the better off than boys. Girls have higher adult sur- index are not perfect proxies of human capital and vival rates in all of the 153 countries for which sex try to balance accuracy and data availability. For disaggregation is available in the Human Capital example, the index does not capture gender bias Index 2020. In all but two countries—India and in terms of sex-selective abortions (what might Tonga—child survival rates are higher for girls be called prebirth survival).30 Moreover, health is than for boys. Meanwhile, girls are more likely proxied by adult mortality rates, but some evi- to be stunted than boys in just 5 of 85 countries: dence shows that, while women live longer than Bhutan, Iraq, Kazakhstan, Moldova, and Tunisia.28 men, they are not necessarily in better health.31 As 27 Bossavie and Kanninen (2018). 28 Stunting rates are calculated using survey data and differences in average rates between girls and boys may not be statistically sig- nificant. 29 United Nations (2011) and Crimmins et al. (2019). 30 The number of “missing women” was estimated to be 126 million in 2010 (Bongaarts and Guilmoto 2015). This refers to the deficit of females relative to males, compared to the figures that would have been observed had all female fetuses been allowed to be born. 31 Guerra et al. (2008); Bora and Saikia (2018). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 21 a measure of the human capital potential of chil- are experiencing prolonged political crises or are dren today, the index does not capture gender gaps undergoing a gradual but still fragile reform and in human capital among the current population of recovery process. These circumstances complicate adults. These caveats are important backdrops to the process of consensus building and resource any analysis of gender gaps using the HCI. Finally, mobilization across political cycles and there- the index implicitly assumes that a child born fore pose a unique set of challenges in improving today will be absorbed into the labor market to use human capital.33 her human capital potential in terms of income generation, when, in fact, female labor participa- The importance of investing in human capital tion rates, globally, are 27 percentage points lower extends beyond the gains it promises in labor than male labor participation rates.32 Chapter 4 productivity and in ensuring that growth is on human capital utilization delves into this, by inclusive and sustainable. It is also a cornerstone proposing an adjustment of the HCI that captures of social cohesion, equity, and trust in institu- labor outcomes. These outcomes reflect one of tions.34 The seven countries scoring lowest on many ways human capital is utilized to improve the Human Capital Index (HCI) in 2020 are all well-being and overall economic development. on the World Bank’s current annual list of fragile and conflict-affected situations (FCS).35 On aver- Equal access to education and health is far from age, countries affected by conflict and violence, realized. Despite progress, girls continue to face compared with the rest of the world, are signifi- greater challenges. Child marriage, household cantly further away from reaching the productiv- responsibilities, teenage pregnancies, and gender-­ ity frontier. based violence in schools pose challenges in keeping girls enrolled, especially, but not only, in Shocks, such as armed conflict or natural disasters, low-income settings. have a lasting impact on human capital. Some pathways for this impact are obvious, including the destruction of human potential through com- 1.5 HUMAN CAPITAL IN FRAGILE AND bat deaths and casualties of natural calamities; CONFLICT-AFFECTED CONTEXTS damage to critical infrastructure and institutions, such as hospitals and schools; and the loss of skills Human capital accumulation requires a sus- resulting from mass displacement. But the impact tained political commitment, an adequate and of these shocks on human capital reach farther. timely resource mobilization, a whole-of-society For instance, emerging evidence shows that the approach, and effective use of data and measure- destructive impacts of armed conflict on health ment. However, these features are not typical of and educational outcomes persist long after the economies that are grappling with fragility, con- fighting stops—extending to future generations flict, and violence. By definition, such settings not yet born when the conflict occurred.36 are plagued with high levels of institutional and social fragility, often with deteriorating gover- Classic studies of conflict and human capital nance capacity. In many cases, these economies have given central attention to health impacts 32 ILOSTAT. Retrieved from World Bank Gender Data Portal. 33 World Bank (2020e). 34 Kim (2018). 35 These countries are the Central African Republic, Chad, Liberia, Mali, Niger, Nigeria and South Sudan. 36 Corral et al. (2020). 22 The Hum an Capita l Index 20 20 U pdate Figure 1.10: Human capital and severity of conflict Human Capital Index Survival rate from Age 15−60 Fraction of Children Under 5 Not Stunted Probability of Survival to Age 5 Learning−Adjusted Years of School Harmonized Test Scores Expected Years of School Lowest Median Highest Average of economies in high−intensity conflict Economy in high−intensity conflict Average of other economies in FCS Other economy in FCS Average of economies not in FCS Economy not in FCS Source: Corral et al. (2020) with updated HCI data for 2020. Notes: FCS = fragile and conflict-affected situations. Economies in high-intensity conflict are defined as having at least 10 conflict deaths per 100,000 population according to the Armed Conflict and Event Data Project (ACLED) and the Uppsala Conflict Data Program (UCDP), while also experiencing a total of more than 250 conflict deaths according to ACLED, or more than 150 conflict deaths according to UCDP. on children exposed to conflict settings. The link had lower weight-for-age and weight-for-height between violent conflict and a range of negative z-scores and higher probability of wasting than health outcomes among children has been estab- children living in less-affected areas.38 lished causally. For example, the physical develop- ment of children who were exposed to the 2002– The intensity of conflict also determines the 07 civil conflict in Côte d’Ivoire was stunted, and extent of human capital depletion. For instance, at this negative impact increased with the length of aggregate level, the distance from the HCI fron- exposure to the conflict. 37 The impact on human tier (an HCI score of 1) increases with the intensity capital increases with increasing conflict sever- of conflict, even among FCS countries. Countries ity. Children living in areas of Nigeria that were with high-intensity conflict, defined as having at heavily affected by the Boko Haram insurgency least 10 conflict deaths per 100,000 people, with 37 Minoiu and Shemyakina (2014). 38 A z-score is a measure of how many standard deviations below or above the population mean a raw score is. See Ekhator-Mobayode and Abebe Asfaw (2019). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 23 Box 1.5: Schooling for Syrian refugee children in Jordan The government of Jordan has adopted a policy of offering refugee children tuition-free access to the public education system, while also providing accredited schools in refugee camps. As a result, overcrowding has occurred in schools in some locations. Despite these measures, access to school for refugee children is still limited. Only about 152,000 of the estimated 236,000 Syrian refugee children present in the country are enrolled (64 percent). Figure 1 paints a stark picture of the enrollment decline by age among refugees. It shows that enrollment significantly tails off after age 11, more so for boys than for girls. This is driven by several factors, including poverty (since most families can’t cover the auxiliary costs of education, such as transportation and school materials); early marriage (which is also evident in recent household surveys); and increased opportunity cost of education, since many children start working early to support their families. Reports suggest that bullying at schools and the absence of a safe learning space impedes learning for Jordanian boys as well as Syrian refugees, and the Jordanian government is taking measures to address this issue. In addition, important reforms such as ensuring universal enrol- ment for 5-year-olds in pre-primary education apply to all inhabitants of Jordan, including Syrian refugee children. Figure 1: Net enrollment rate of Syrian refugees in formal education in Jordan 100 80 60 40 20 0 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Age Girls Boys Source: Krafft et al. (2018). Notes: Data only include refugees registered with the UN High Commissioner for Refugees (UNHCR). a minimum of 150 casualties, have consistently outcomes among women in Nepal exposed as scored lower on all components of the index, children to the country’s post-1996 civil war were compared with other FCS and non-FCS econo- significantly worse than for those women and chil- mies (Figure 1.10). dren who were not exposed to conflict. Not only did the first-generation victims show significant Conflict can have adverse effects on human cap- reductions in final adult height, but when those ital across generations. Well-being and health conflict-exposed victims had children of their own, 24 The Hum an Capita l Index 20 20 U pdate their children also suffered reduced weight-for- The intergenerational impact of conflict and vio- height and body mass index z-scores, on average. lence extends to losses in educational attainment Women exposed to the conflict during childhood for children not even born when fighting took had more children and lived in poorer households place. For example, exposure to the Rwandan as adults. The combination of these two factors genocide in utero decreased educational attain- may decrease parents’ ability to invest in their ment by 0.3 years and the likelihood of complet- children’s human capital during critical phases of ing primary school by 8 percent.43 The impact on physical and cognitive development and therefore years of schooling was stronger for females and propagate these impacts intergenerationally.39 for individuals exposed to the genocide in the first trimester of gestation. Each additional month of Human capital depletion in FCS countries also hap- exposure in utero decreased educational attain- pens through reduced and unequal access to educa- ment by 0.21 years of schooling. Through in utero tion and poor learning outcomes among those who exposure, conflict-related disruptions of fetal cog- do have access. Refugee and internally displaced nitive development may affect children’s subse- children embody the losses of educational human quent cognitive capacities, educational outcomes, capital associated with armed conflict. Conflict in and earning power as adults. Syria, for example, has led to disruptions in educa- tion for millions of children, including over a million How can fragile countries and development part- who have been forced to flee to neighboring coun- ners confront the losses of human capital driven tries.40 Jordan hosts one of the largest populations by conflict? The best solution is to prevent fragility, of Syrian refuges and has made concerted effort to conflict, and violence from engulfing countries in provide access to education for refugee children. the first place. But when conflict does erupt, effective Yet, Syrian refugee children in Jordan experience delivery of health and education services tailored to delayed entry into school and early exit, with enroll- FCS conditions is vital. Only the preservation and ment rates dropping sharply from around age 12, as rebuilding of human capital can enable countries to refugee children come under pressure to work and durably escape cycles of fragility and violence. help support their families (Box 1.5).41 Yet, the delivery of health and education ser- Globally, refugees access education at much lower vices in FCS poses daunting challenges, not least rates than other children. In 2016, only 61 per- because of the extreme diversity of FCS contexts. cent of refugee children attended primary school, However, much has recently been learned from compared to 91 percent of all children. At the the experiences of various countries. secondary level, 23 percent of refugee children are enrolled, versus 84 percent of eligible young In Afghanistan, for example, following the with- people worldwide.42 These shortfalls are especially drawal of the Taliban in 2001, the Ministry of concerning, because the number of refugees and Public Health (MOPH) had to provide emergency displaced people worldwide has risen steadily relief services to address the grave health situation through the past decade and now stands at its throughout the country. Yet the health system was highest level since World War II. in ruins after decades of warfare and neglect. As 39 Phadera (2019). 40 Sieverding et al. (2018). 41 Tiltnes, Zhang, and Pedersen (2019). 42 UN High Commissioner for Refugees (UNHCR 2017). 43 Bundervoet and Fransen (2018). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 25 they rolled out emergency health services, includ- learning, life skills, and social cohesion. Only by ing in many areas still subject to conflict, health securing broad dissemination of these capac- officials had to plan for the future, which included ities in the population through quality educa- rebuilding and sustaining a functional national tion can countries build lasting foundations for health system. Acknowledging its capacity limita- peace and economic recovery. But the challenges tions and with technical assistance from the inter- in delivering equitable, quality education is not national community, the MOPH led the creation straightforward. of an innovative public-private partnership frame- work for health service delivery in Afghanistan.44 Some of the greatest service delivery challenges This delivery model has improved key health indi- in conflict-affected countries involve education cators under highly challenging conditions and for displaced populations and host communities. has been recognized as an example for other post- Over the years, various flexible learning strate- conflict countries.45 gies have been fielded across different settings. Learning from these global experiences, there is Without adequate health financing, health ser- a consistent move away from providing refugee vice delivery simply will not happen. The ongoing education in parallel systems that may lack quali- Syrian crisis has underscored that country authori- fied teachers, consistent funding, ability to provide ties, donors, and international partners must coor- diplomas, and quality control. Ethiopia’s Refugee dinate their efforts to ensure that health and other Proclamation, for example, gives refugees access to essential services for refugees can be sustainably national schools and gives host children access to paid for. An important resource to facilitate such refugee schools. Iran decreed in 2015 that schools durable support came with the 2016 launch of the accept all Afghan children regardless of documen- Global Concessional Financing Facility (GCFF). tation. Turkey has committed to include all Syrian Led jointly by the World Bank, the United Nations, refugee children in its national education system and the Islamic Development Bank, the GCFF is a by 2020.46 The inclusion of refugees in national global platform designed to deliver concessional education systems dramatically expands educa- funding to middle-income countries that provide tional opportunities for refugees. But the process a global public good by hosting large numbers of remains fraught with challenges related to system refugees. GCFF resources enable governments capacities, persistent access barriers, quality, and in host countries to offer expanded services to resources. Furthermore, the refugees, while continuing to meet the needs of their own citizens. Early GCFF concessional loans The lack of timely, reliable, and actionable data reduced the acute financial burden on Lebanon and a robust measurement agenda are also factors and Jordan, two countries on the front lines of the hindering progress in human capital accumula- Syrian refugee response. Subsequently, the GCFF tion in FCS economies. Although high-quality has worked to smooth the transition from human- data are critical for diagnosing deficiencies and itarian assistance to development by providing formulating targeted policies and programs to medium- and long-term concessional finance. enhance human capital, such data are not read- ily available across many countries that are in the Even more than other countries, economies in midst of, or recovering from, fragility and con- FCS need education systems that can promote flict. For instance, for some of the countries that 44 Newbrander, Waldman, and Shepherd-Banigan (2011). 45 World Bank (2018b). 46 UNESCO (2019). 26 The Hum an Capita l Index 20 20 U pdate are classified as FCS as per the World Bank 2020 address the lack of a sampling frame in Democratic classification, the HCI score cannot be calculated. Republic of Congo and Somalia. When data collec- This may be because data informing various HCI tion is hampered by security concerns for enumer- components either do not exist or are outdated. ators, locally recruited, resident enumerators can Even when the index can be calculated compatibly make it possible to collect relevant, reliable, and with the HCI data inclusion rules (see appendix A), timely evidence that can shed light on the plight it might still not fully capture the deterioration of of the most vulnerable populations. These efforts human capital that can follow the rapidly chang- can move the needle on addressing persistent data ing reality of countries in conflict. In the case deprivation in FCS economies. of Yemen, available data for index components mostly predate the conflict and might not fully Protecting and rebuilding human capital in set- represent the effect of the conflict on schooling or tings of fragility and conflict are crucial to restore child health. Worse still, comparable data for refu- hope in these countries. They are also critical gees and hosts is almost non-existent in countries for reaching global poverty goals. Over recent afflicted by fragility and conflict. decades, poverty has become steadily more con- centrated in economies in FCS. Fragility and con- Collecting high-quality data requires sustained and flict deplete human capital. Yet societies must rely deliberate efforts. In light of other pressures in sit- heavily on human capital to recover from fragility uations of violence and conflict, measurement is and conflict. This paradox underscores the impor- rarely a priority. However, collecting high-quality tance of health and education services in FCS data that address stakeholders’ data needs is feasible settings. Delivering these services lays the foun- in these settings. For instance, mobile phone inter- dations that will enable countries to emerge from views have been used for data collection during cycles of violence and return to peace, stability, the Ebola crisis in Sierra Leone and to inform a and development. Overcoming systemic barriers response to drought in Nigeria, Somalia, South will simultaneously require careful coordination Sudan, and Yemen. Likewise, satellite images and 47 between humanitarian and development partners machine learning algorithms were employed to and a whole-of-society approach. 47 Hoogeveen and Pape (2020). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 27 1.6 HCI 2020 UPDATE TABLE Table 1.2: The Human Capital Index (HCI), 2020 Lower Upper Lower Upper Lower Upper Economy Bound Value Bound Economy Bound Value Bound Economy Bound Value Bound Central African Republic 0.26 0.29 0.32 India 0.49 0.49 0.50 Mauritius 0.60 0.62 0.64 Chad 0.28 0.30 0.32 Egypt, Arab Rep. 0.48 0.49 0.51 Uzbekistan 0.60 0.62 0.64 South Sudan 0.27 0.31 0.33 Guyana 0.48 0.50 0.51 Brunei Darussalam 0.62 0.63 0.63 Niger 0.29 0.32 0.33 Panama 0.49 0.50 0.51 Kazakhstan 0.62 0.63 0.63 Mali 0.31 0.32 0.33 Dominican Republic 0.49 0.50 0.52 Costa Rica 0.62 0.63 0.64 Liberia 0.30 0.32 0.33 Morocco 0.49 0.50 0.51 Ukraine 0.62 0.63 0.64 Nigeria 0.33 0.36 0.38 Tajikistan 0.48 0.50 0.53 Seychelles 0.61 0.63 0.66 Mozambique 0.34 0.36 0.38 Nepal 0.49 0.50 0.52 Montenegro 0.62 0.63 0.64 Angola 0.33 0.36 0.39 Micronesia, Fed. Sts. 0.47 0.51 0.53 Albania 0.62 0.63 0.64 Sierra Leone 0.35 0.36 0.38 Nicaragua 0.50 0.51 0.52 Qatar 0.63 0.64 0.64 Congo, Dem. Rep. 0.34 0.37 0.38 Nauru 0.49 0.51 0.53 Turkey 0.64 0.65 0.66 Guinea 0.35 0.37 0.39 Fiji 0.50 0.51 0.52 Chile 0.64 0.65 0.66 Eswatini 0.35 0.37 0.39 Lebanon 0.50 0.52 0.52 Bahrain 0.64 0.65 0.66 Yemen, Rep. 0.35 0.37 0.39 Philippines 0.50 0.52 0.53 China 0.64 0.65 0.67 Sudan 0.36 0.38 0.39 Tunisia 0.51 0.52 0.52 Slovak Republic 0.66 0.66 0.67 Rwanda 0.36 0.38 0.39 Paraguay 0.51 0.53 0.54 United Arab Emirates 0.66 0.67 0.68 Côte d'Ivoire 0.36 0.38 0.40 Tonga 0.51 0.53 0.55 Serbia 0.67 0.68 0.69 Mauritania 0.35 0.38 0.41 St. Vincent and the Grenadines 0.52 0.53 0.54 Russian Federation 0.67 0.68 0.69 Ethiopia 0.37 0.38 0.39 Algeria 0.53 0.53 0.54 Hungary 0.67 0.68 0.69 Burkina Faso 0.36 0.38 0.40 Jamaica 0.52 0.53 0.55 Luxembourg 0.68 0.69 0.69 Uganda 0.37 0.38 0.40 Indonesia 0.53 0.54 0.55 Vietnam 0.67 0.69 0.71 Burundi 0.36 0.39 0.41 Dominica 0.53 0.54 0.56 Greece 0.68 0.69 0.70 Tanzania 0.38 0.39 0.40 El Salvador 0.53 0.55 0.56 Belarus 0.69 0.70 0.71 Madagascar 0.37 0.39 0.41 Kenya 0.53 0.55 0.56 United States 0.69 0.70 0.71 Zambia 0.38 0.40 0.41 Samoa 0.54 0.55 0.56 Lithuania 0.70 0.71 0.72 Cameroon 0.38 0.40 0.42 Brazil 0.55 0.55 0.56 Latvia 0.69 0.71 0.72 Afghanistan 0.39 0.40 0.41 Jordan 0.54 0.55 0.56 Malta 0.70 0.71 0.72 Benin 0.38 0.40 0.42 North Macedonia 0.55 0.56 0.56 Croatia 0.70 0.71 0.72 Lesotho 0.38 0.40 0.42 Kuwait 0.55 0.56 0.57 Italy 0.72 0.73 0.74 Comoros 0.36 0.40 0.43 Grenada 0.55 0.57 0.58 Spain 0.72 0.73 0.73 Pakistan 0.39 0.41 0.42 Kosovo 0.56 0.57 0.57 Israel 0.72 0.73 0.74 Iraq 0.40 0.41 0.41 Georgia 0.56 0.57 0.58 Iceland 0.74 0.75 0.75 Malawi 0.40 0.41 0.43 Saudi Arabia 0.56 0.58 0.59 Austria 0.74 0.75 0.76 Botswana 0.39 0.41 0.43 Azerbaijan 0.56 0.58 0.59 Germany 0.74 0.75 0.76 Congo, Rep. 0.39 0.42 0.44 Armenia 0.57 0.58 0.59 Czech Republic 0.74 0.75 0.76 Solomon Islands 0.41 0.42 0.43 Bosnia and Herzegovina 0.57 0.58 0.59 Poland 0.74 0.75 0.76 Senegal 0.40 0.42 0.43 West Bank and Gaza 0.57 0.58 0.59 Denmark 0.75 0.76 0.76 Gambia, The 0.39 0.42 0.44 Moldova 0.57 0.58 0.59 Cyprus 0.75 0.76 0.76 Marshall Islands, Rep. 0.40 0.42 0.44 Romania 0.57 0.58 0.60 Switzerland 0.75 0.76 0.77 South Africa 0.41 0.43 0.44 St. Kitts and Nevis 0.57 0.59 0.60 Belgium 0.75 0.76 0.77 Papua New Guinea 0.41 0.43 0.44 Palau 0.57 0.59 0.61 France 0.75 0.76 0.77 Togo 0.41 0.43 0.45 Iran, Islamic Rep. 0.58 0.59 0.60 Portugal 0.76 0.77 0.78 Namibia 0.42 0.45 0.47 Ecuador 0.59 0.59 0.60 Australia 0.76 0.77 0.78 Haiti 0.43 0.45 0.46 Antigua and Barbuda 0.58 0.60 0.61 Norway 0.76 0.77 0.78 Tuvalu 0.43 0.45 0.46 Kyrgyz Republic 0.59 0.60 0.61 Slovenia 0.77 0.77 0.78 Ghana 0.44 0.45 0.46 Sri Lanka 0.59 0.60 0.60 New Zealand 0.77 0.78 0.78 Timor-Leste 0.43 0.45 0.47 Uruguay 0.59 0.60 0.61 Estonia 0.77 0.78 0.79 Vanuatu 0.44 0.45 0.47 Argentina 0.59 0.60 0.61 United Kingdom 0.77 0.78 0.79 Lao PDR 0.44 0.46 0.47 St. Lucia 0.59 0.60 0.62 Netherlands 0.78 0.79 0.80 Gabon 0.43 0.46 0.48 Trinidad and Tobago 0.57 0.60 0.62 Ireland 0.78 0.79 0.80 Guatemala 0.45 0.46 0.47 Colombia 0.59 0.60 0.62 Sweden 0.79 0.80 0.81 Bangladesh 0.46 0.46 0.47 Peru 0.59 0.61 0.62 Macao SAR, China 0.79 0.80 0.80 Zimbabwe 0.44 0.47 0.49 Oman 0.60 0.61 0.62 Finland 0.79 0.80 0.80 Bhutan 0.45 0.48 0.50 Thailand 0.60 0.61 0.62 Canada 0.79 0.80 0.81 Myanmar 0.46 0.48 0.49 Malaysia 0.60 0.61 0.62 Korea, Rep. 0.79 0.80 0.81 Honduras 0.47 0.48 0.49 Mexico 0.60 0.61 0.62 Japan 0.80 0.80 0.81 Cambodia 0.47 0.49 0.51 Bulgaria 0.60 0.61 0.62 Hong Kong SAR, China 0.80 0.81 0.82 Kiribati 0.46 0.49 0.52 Mongolia 0.60 0.61 0.63 Singapore 0.87 0.88 0.89 HCI < 0.40 0.40 ≤ HCI < 0.50 0.50 ≤ HCI < 0.60 0.60 ≤ HCI < 0.70 0.70 ≤ HCI < 0.80 0.80 ≤ HCI Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The Human Capital Index ranges between 0 and 1. The index is measured in terms of the productivity of the next generation of workers relative to the benchmark of complete education and full health. An economy in which a child born today can expect to achieve complete education and full health will score a value of 1 on the index. Lower and upper bounds indicate the range of uncertainty around the value of the HCI for each economy. 28 The Hum an Capita l Index 20 20 U pdate Box 1.6: Where did the HCI rankings go? The 2020 update does not report rankings for the 174 countries with an HCI score. There are four reasons for this. First, coverage of the index has increased by 17 countries, from 157 countries in the inaugural 2018 HCI to 174 countries in 2020. So, for instance, a rank of 37 out of 157 cannot be compared with a rank of 37 out of 174. Given the change in HCI coverage between 2018 and 2020, simple comparisons of rankings as an indication of a country’s progress over time are meaningless. Second, even if comparisons were restricted to the set of countries that are part of both the 2018 and 2020 versions of the index, rankings artificially inflate small differences in HCI scores. For example, there are eight countries clustered between HCI scores of 0.60 and 0.61, so if one of those countries at 0.60 improves by just 0.01, it would move up eight places in the ranking. By contrast, there are just two countries between 0.70 and 0.71, and so if one of those two countries were to improve its score by 0.01, it would only move up one rank.a Third, rankings suppress information on the absolute gains and losses countries have made on the HCI. Consider for example the comparison of HCI 2020 and HCI 2010, which is graphed in the left panel of Figure 1.12.b Most countries have improved their human capital outcomes, reflected by the fact that they are above the 45-degree line in the figure. Rankings cannot con- vey these gains (or losses), because they only present the positions of countries relative to each other. This is illustrated in the right panel of Figure 1, which plots the same information for 2020 versus 2010 but in rank terms. Even countries that have made gains in human capital accumula- tion may fall below the 45-degree line simply because of their position relative to other countries. In addition, points in the right panel are more spread out compared with those on the left, illus- trating how ranks artificially magnify small changes, as stated above. Figure 1: Changes in HCI scores and ranks, 2010-2020 0.9 100 Finland Macao SAR, China Macao SAR, China Finland Switzerland Cyprus 80 Human Capital Index, circa 2020 Cyprus Switzerland Italy Italy 0.7 Greece HCI rank, circa 2020 Russian Federation Greece Slovak Republic Russian Federation 60 Slovak Republic Albania Albania Bulgaria Ecuador Azerbaijan Romania Bulgaria 40 Ecuador Romania 0.5 Panama Azerbaijan Panama 20 Côte d’Ivoire Côte d’Ivoire 0.3 0 0.3 0.5 0.7 0.9 0 20 40 60 80 100 Human Capital Index, circa 2010 HCI rank, circa 2010 EAP ECA LAC MENA NA SAR SSA T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 29 Fourth, and most important, there is no need to focus on country ranks because the index itself is expressed in units that are meaningful. Because the HCI is measured in terms of the produc- tivity of the next generation of workers relative to the benchmark of complete education and full health, the units of the index have a natural interpretation: a value of 0.50 for a country means that the productivity as a future worker of a child born in a given year in that country is only half of what it could be under the benchmark. Rankings place an inordinately large focus on the fact that a country with an HCI of 0.51 is ahead of a country with an HCI of 0.50. But this interpretation misses the more critical issue, which is that in both countries, children born today will grow up with half their human capital potential unfulfilled. This is vastly more important than whether one country is “ahead of” another. a This problem is amplified by the fact that the components of the HCI are measured with some error, and this uncertainty naturally has implications for the precision of the overall HCI. To capture this imprecision, the HCI estimates for each country are accompanied by upper and lower bounds that reflect the uncertainty in the measurement of HCI components. In cases where these intervals overlap for two countries, this indicates that the differences in the HCI estimates for these two countries should not be overinterpreted, because they are small relative to the uncertainty around the value of the index itself. Rankings further amplify these minor differences. b The construction of an HCI for 2010 is described in the following chapter. 2 Human Capital Accumulation OVER TIME T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 31 A s the first update of the HCI, the 2020 release 2.1 HUMAN CAPITAL ACCUMULATION is an opportunity to look at the evolution of OVER THE PAST DECADE human capital outcomes, as measured by the HCI, across countries over time. To track progress over the past decade, a version of the HCI has been calculated for 103 economies Unlike indices that aggregate laws or regulation, using component data in or near 2010. which can be modified by swift government leg- islative or regulatory action, the HCI is based on Data used to populate the 2010 HCI have been outcomes that typically change slowly from year carefully selected to maximize comparability with to year. Some of them—such as stunting and edu- the 2020 HCI. In particular, only those countries cational test scores—are measured infrequently, where learning scores were measured by the same every three to five years. As a result, changes in international assessment in 2010 and 2020 enter the HCI over a short period are small and might the comparison (see Box 2.1). This requirement for simply reflect updates to components that are test scores from the same testing program proved measured sporadically. In contrast, the analy- the main constraint to building large and repre- sis of longer-term trends has a more solid basis, sentative coverage, and the resulting 2010 sample given the scope for smoothing out short-run is, unsurprisingly, skewed toward richer countries idiosyncrasies. that tend to have more complete and better qual- ity data. For example, the sample does not cover This chapter examines trends in human capital South Asia, since none of the seven countries in over time. Section 1 discusses the construction the region with an HCI in 2020 have learning data of an HCI for 2010 and the evolution of the HCI in 2010 from the same representative interna- between 2010 and 2020. Section 2 unpacks these tional test assessment as in 2020. The average HCI dynamics by looking at changes in the compo- in 2020 for those economies that also have an HCI nents of the HCI. Section 3 further sharpens in 2010 is 0.62, while it is 0.48 for economies not the country and policy focus. Drawing lessons included in the sample (Table 2.1). The potential from case studies, it shows that a longer-run bias is largest in the East Asia and Pacific region, perspective on country trajectories can high- where the HCI is 40 percent higher in countries light promising policies, including the role that a with a 2010 HCI than in those without (gross whole-of-society approach, steady political com- domestic product [GDP] per capita is 88 percent mitment, domestic resource mobilization, and higher). evidence-based policies can play in human capi- tal progress. As measured by the HCI, human capital improved in most countries in the last decade. Figure 2.1 plots HCI 2020 scores against HCI 2010 scores, likely reflecting underlying secular trends in various dimensions of human capital. On average, the HCI increased by 2.6 points between 2010 and 2020. 32 Hum an Capital Accumu lati on over T i m e Table 2.1: Regional coverage, HCI 2020 and HCI 2010 ECONOMIES WITH A 2020 HCI ECONOMIES WITH A 2010 HCI Real GDP Number of Real GDP Number of REGION HCI 2020 per capita economies HCI 2020 per capita economies East Asia and Pacific 0.59 23,376 31 0.71 43,977 12 Europe and Central Asia 0.69 35,278 48 0.71 39,479 41 Latin America and the Caribbean 0.56 15,572 26 0.58 18,444 13 Middle East and North Africa 0.57 28,437 18 0.60 34202 14 North America 0.75 55,857 2 0.75 55,857 2 South Asia 0.48 6,605 7 — — — Sub-Saharan Africa 0.40 5,125a 42 0.42 6,586 21 Average, total 0.56 21,403a 174 0.62 30,243 103 Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data and the World Development Indicators and Penn World Tables 9.1 for per capita GDP data. Notes: The table uses real GDP per capita at PPP, in constant 2011 USD, for most recently available data as of 2019; — = data not available. Per capita GDP data for South Sudan are not available. a Figure 2.1: Changes in the Human Capital Index, circa 2010–circa 2020 1.0 Macao SAR, China 0.8 Human Capital Index, circa 2020 Finland Cyprus Switzerland EAP Italy Greece ECA Russian Federation Slovak Republic LAC Albania Bulgaria 0.6 Ecuador MENA Azerbaijan Romania NA Zimbabwe Panama SAR Togo SSA 0.4 Lesotho Côte d’Ivoire 0.2 0.2 0.4 0.6 0.8 Human Capital Index, circa 2010 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the 2020 HCI (on the vertical axis) against the 2010 HCI (on the horizontal axis) for 103 countries where data are available for both 2010 and 2020. The dashed line is a 45-degree line. Points above (below) represent an increase (decrease) in the HCI between 2010 and 2020. For countries in which the HCI scores improved— that experienced a rise in the index had increases about 80 percent of the sample, depicted above above 5 points. This means that, in those countries, the 45-degree line in Figure 2.1—scores increased the productivity of future workers approached by an average of 3.5 points. One economy in four the frontier by 5 percentage points—a substantial T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 33 Box 2.1: Ensuring comparability across time in the HCI The 2020 update of the HCI also reports a version of the HCI calculated for 2010, offering an opportunity to track progress on human capital outcomes. The outcome measures that are used to calculate the HCI typically register only small changes from one year to the next. A time frame of 10 years allows the index to track real underlying change in human capital outcomes over a longer period, smoothing out short-run idiosyncrasies. The HCI for 2010 is calculated for 103 countries where comparable data are available, and it provides a benchmark year for countries to measure changes over time as well as the pace of their progress. The data used to populate the 2010 HCI are selected to be “near” 2010 and to maximize comparability with 2020. This is straightforward in the case of child survival rates that are updated annually and adult survival rates that are updated every two years.a While enroll- ment rates used to calculate the EYS are reported annually for some countries, others may have significant gaps in their time series. In the case of gaps in enrollment for 2010, data are imputed using an annualized growth rate derived from available enrollment data for the country.b In the case of more sporadically reported stunting and test scores data, the surveys and tests used to populate the two time periods are selected to typically be at least five years apart and as close as possible to 2010 and 2020. In the case of test scores, an additional requirement that both data points come from the same or a highly comparable testing assessment program ensures comparability over time. The five exceptions are Algeria, where harmonized test scores from the Progress in International Reading Literacy Study (PIRLS) or the Trends in International Mathematics and Science Study (TIMSS) in 2007 are used to populate the 2010 HCI, while harmonized test scores based on the Program for International Student Assessment (PISA) in 2015 are used to populate the 2020 HCI; and Morocco, North Macedonia, Saudi Arabia, and Ukraine, where data from PIRLS or TIMSS in 2011 are used for the 2010 HCI, while data from PISA 2018 are used for the 2020 HCI. To maximize comparability with PISA, only secondary-level scores from TIMSS and PIRLS are used to calculate the 2010 HTS for these five countries. Finally, while child survival, EYS, and harmonized test scores are essential to calculating an HCI, the fraction of children not stunted and adult survival rate both act as proxies for latent health. Consequently, the HCI can be calculated using either one of these proxies if both are not available.c To ensure comparability in HCI scores over time, the same health proxies are used to calculate both the 2010 and 2020 scores. This means that if data for stunting are unavailable in 2010, they are not used to calculate the HCI for 2020, and vice versa. a Adult survival rates are the complement of mortality rates for 15– through 60-year-olds, reported for five-year periods by the United Nations Population Division. These data are linearly interpolated to produce the annual estimates for countries used to calculate the HCI. See the section on Adult Mortality Rates in Appendix B for more details. b The methodology to fill in gaps in enrollment data is described in detail in the Expected Years of School section of Appendix B. c See Appendix A for a detailed description of how HCI components are aggregated to calculate the final index. 34 Hum an Capital Accumu lati on over T i m e progress. Over time, there is convergence in the low-income countries, human capital improved Human Capital Index. That is, in economies start- slightly more quickly relative to GDP per capita. ing at lower values of the HCI in 2010, human With health accounting for an important share of capital improved more rapidly than in economies improvement in the index, especially in low-in- where the HCI was higher to being with, even after come countries (see section 2.2), a steeper slope accounting for initial GDP per-capita. 1 in the HCI-GDP relationship likely reflects global gains in health, such as better and less expen- The economies with the largest gains include sive treatments and improved technology, which the Macao Special Administrative Region of the benefited all countries but brought about larger People’s Republic of China, Albania, the Russian advances in poorer countries. Federation, Côte d’Ivoire, and Azerbaijan. A vari- ety of factors account for these improvements: Regional and income group averages mask signif- improved learning as measured by higher test icant differences in individual country trajecto- scores (Macao SAR, China, Albania), better health ries, which are depicted in Figure 2.2, panel c. For (in the case of Russia, specifically improvements example, in Azerbaijan, human capital outcomes in adult survival, marking a rebound from the increased by 0.08 (from 0.50 to 0.58), while there drop in life expectancies in the post-Soviet era)2, was almost no change in the country’s GDP per and school enrollment (at the pre-primary level capita. By contrast, Lithuania experienced only in Azerbaijan, at the primary level in Macao SAR, a small increase in the HCI despite a significant China, and Côte d’Ivoire, at the secondary level in increase in per capita income.. the Russian Federation). Looking back over the last decade shows that both Some countries experienced modest declines in girls and boys have made strides in improving the index. These include the Republic of Korea, human capital. Sex-disaggregated data are avail- Greece, Bulgaria, and Italy, where the index fell by able for 90 countries in the comparison sample about 2 HCI or more points. Among the 10 coun- over time (Figure 2.3). The average gender ratio tries with the largest drops, eight are European is similar in circa 2010 and circa 2020, at about countries, and only one of the 10 is not a high-in- 1.06 in favor of girls. However, this stable aver- come economy. These decreases in the HCI can be age masks considerable differences at the country mainly traced back to drops in test scores. level. Around 2010, in all but seven economies, the HCI was higher among girls than among boys. As incomes increase, on average, human capital Among the seven economies where girls were at improves. Panels a and b of Figure 2.2 indicate a disadvantage, the girl-boy ratio improved in the direction of change of the HCI from 2010 to Cameroon, Chad, and Côte d’Ivoire but did not 2020, denoted by the dots and the arrow points reach full gender parity in the last decade. These respectively. The slopes of the arrows signal the are the countries in the lower left quadrant of rate at which rising per capita income is associated Figure 2.3, above the 45-degree line and below the with more human capital. The pace is quite uni- horizontal red line. Meanwhile, in Benin, Burkina form across country income groups. However, in Faso, and Morocco, girls fully caught up with boys, 1 All the components of the HCI, and the HCI itself, are bounded above. For example, adult and child survival rates cannot be larger than 100 percent, and the maximum number of learning-adjusted years of school between ages 4 and 17 is fixed at 14. This means that the absolute size of improvements become smaller as countries get closer to the upper bound. 2 Smith and Nguyen (2013). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 35 Figure 2.2: Human capital and GDP per capita: Changes over time a. Changes by income group b. Changes by regional group 1.0 1.0 0.9 0.9 Human Capital Index Human Capital Index 0.8 0.8 NA 0.7 0.7 HIC EAP ECA 0.6 0.6 LAC MENA UMIC 0.5 0.5 LMIC 0.4 0.4 SSA LIC 0.3 0.3 6 8 10 12 6 8 10 12 Log GDP Per Capita Log GDP Per Capita c. Changes at the country level 1.0 Japan 0.8 Human Capital Index Vietnam Serbia Macao SAR, China 0.6 Qatar Ecuador Oman Zimbabwe Azerbaijan Guatemala South Africa 0.4 Togo Burundi Côte d’Ivoire 0.2 6 8 10 12 Log GDP Per Capita Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: Panel A (panel B) plots the average HCI for income groups (regional groups) using the World Bank Group classification (on the vertical axis) against log real GDP per capita (on the horizontal axis) for 103 countries where data are available for both 2010 and 2020. The 2010 HCI is denoted by dots and the HCI 2020 is denoted by an arrow. Panel C plots country-level data for HCI 2010 and HCI 2020 (on the vertical axis), represented by dots and arrows respectively, against log real GDP per capita (on the horizontal axis) for the 103 countries where data are available for both 2010 and 2020. LIC = low-income countries; LMIC = lower-middle-income countries; UMIC = upper-middle-income countries; HIC = high-income countries. even surpassing them in the latter two countries.3 female-male ratio in the HCI does not capture Among the 83 countries in which the HCI was gaps in other areas of human capital development, higher for girls in 2010, the ratio in favor of girls such as labor force participation. In many coun- had widened in 34 countries. However, a favorable tries, women participate in the labor force at far 3 In Togo, the gender gap in HCI slightly widened in favor of boys. Both boys and girls’ outcomes were improved during this time period. The widening gender gap is driven by different rates of improvement among boys and girls. In expected years of school, girls’ outcomes improved but by a slightly lesser amount compared with boys. Meanwhile, in infant survival and stunting, boys are catching up to girls, closing the gender gaps towards parity. 36 Hum an Capital Accumu lati on over T i m e Figure 2.3: Ratio of women to men, HCI 2010–20 1.20 Georgia 1.15 Lithuania Female−Male Ratio in HCI, circa 2020 Vietnam Low income 1.10 South Africa Burundi Estonia Lower-middle income Timor−Leste Kazakhstan Upper-middle income 1.05 United Arab Emirates High income Congo, Rep. Argentina Montenegro Cyprus 1.00 Cameroon Côte d’Ivoire Peru 0.95 Chad 0.95 1.00 1.05 1.10 1.15 1.20 Female−Male Ratio in HCI, circa 2010 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the gender ratio (female to male) of the 2020 HCI (on the vertical axis) against the 2010 HCI (on the horizontal axis) for 90 countries where sex-disaggregated data are available for both 2010 and 2020. The grey dashed line is a 45-degree line, points above (below) represent an increase (decrease) in the HCI gender ratio between 2010 and 2020. The red horizontal (vertical) line indicates gender parity in the HCI in 2020 (2010). lower rates than men. This point is taken up fur- for the HCI 2020 cross-section, a decomposition4 ther in chapter 4, which discusses an extension of suggests that almost one-third of changes in the the HCI capturing labor market utilization. HCI over the past decade are due to gains in health, as proxied by reductions in stunting and improve- ments in adult survival. Considered together, prog- 2.2 CHANGES IN KEY HUMAN CAPITAL ress in child survival, stunting, and adult survival DIMENSIONS IN THE PAST DECADE accounts for close to half the increase in the HCI; the remainder is explained by changes in enroll- 2.2.1 Component contributions to changes ment and, to a limited extent, learning (Figure 2.4). in the HCI The evolution of the HCI reflects changes in the While countries in every income group expe- components of the index. There are considerable rienced an increase in their HCI, the factors that differences in the pace of change across compo- contributed to these improvements differ across nents and in the extent to which they contribute to income groups, reflecting both countries’ initial changes in the overall HCI. Similar to the analysis conditions and their development trajectories. 4 This decomposition is implemented as a Shapley decomposition. For a description of the method see Azevedo, Sanfelice, and Nguyen (2012). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 37 Figure 2.4: Component contribution to HCI gains, 2010–20 1.00 0.90 HCI points contributed (x100) 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 Child mortality Pre−primary Primary Lower-secondary Upper-secondary Harmonized Test Health enrollment enrollment enrollment enrollment Scores Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: This figure reports a decomposition computed for 103 countries where data are available for both 2010 and 2020. Low-income countries, in the sample, experienced While the HCI comparison between 2010 and considerable gains in child survival rates (which, 2020 is only possible for 103 countries, compar- on average, rose from 90.6 percent in 2010 to 93.4 isons between these two points in time for each percent in 2020). Low-income countries also reg- of the HCI components are possible for a larger istered growth in enrollment rates in pre-primary (and variable) number of countries. The analysis education (from 26.6 to 42.5 percent) and at the that follows includes all countries for which data primary level (from 82.3 to 89.6 percent). These are available, in order to provide a comprehensive gains were offset in some countries by declines picture of changes in the different dimensions of in measured learning. In high-income countries, human capital. The specific trajectories of indi- which were already closer to the frontier for most vidual components are discussed below. components, increases in the HCI are mostly explained by gains in upper-secondary enrollment Child survival and improvements in health, as proxied by adult Progress in child survival over the past decade has survival (Figure 2.5). been substantial in many countries, improving in 136 of the 173 countries for which data are avail- 2.2.2 Changes in index components over time able, as depicted in Figure 2.6.5 On average, the The analysis in this subsection considers the evo- child survival rate rose from 0.96 to 0.97, which lution of components of the HCI over the last translates to 10 fewer deaths per 1,000 live births.6 decade. On average, there has been progress on At an average of 3.6 percentage points, improve- most components of the HCI, as illustrated in ments have been most significant among low-in- Table 2.2, which looks at the sample of countries come countries, which started out with lower rates. with an index in 2020 and 2010. In countries such as Angola, Malawi, Niger, Sierra 5 The child (under-5) mortality rate rose in Grenada, Mauritius, Fiji, Brunei, and Dominica, reported here in ascending order of the increase. 6 See United Nations Interagency Group for Child Mortality Estimation (UNIGME 2019). 38 Hum an Capital Accumu lati on over T i m e Figure 2.5: Contribution to changes in the HCI, by country-income group, 2010–20 Child mortality Pre−primary enrolment Primary enrollment Lower-secondary enrollment Upper-secondary enrollment Harmonized Test Scores Health −0.10 −0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 Share of change explained Low income Lower-middle Upper-middle High income Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: This figure reports a decomposition computed for 103 countries where data are available for both 2010 and 2020. Table 2.2: Changes in HCI components, 2020 from 2010 Component Global EAP ECA LAC MENA NA SSA Survival, percentage point difference 0.007 0.004 0.002 0.004 0.003 0.001 0.022 EYS (year difference) 0.437 0.651 0.176 0.351 0.458 0.440 0.862 HTS (score difference) -0.110 -3.659 1.002 8.001 0.443 -5.769 -5.106 Fraction of Children Under 5 Not Stunted, 0.056 0.048 0.048 0.051 0.034 — 0.070 percentage point difference Adult Survival Rate, percentage point 0.030 0.013 0.020 0.013 0.015 0.002 0.082 difference Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: This table reports changes in regional averages (as defined by the World Bank Group regional classification), computed for 103 countries where data are available for both 2010 and 2020. EYS = expected years of school; HTS = harmonized test score; — = data unavailable. Leone, and Zimbabwe, improvements in child These improvements are the result of global survival meant between 39 and 58 fewer deaths improvements in health but also of a combination per 1,000 live births.7 of extension of health coverage, better maternal and childcare, and better sanitation. For example, 7 A unique case is Haiti, where the child survival rate dropped massively and abruptly to 79 percent (79 of 100 children survive) in 2010 from 92 percent in 2009, following a major earthquake. Survival rates have since rebounded to 94 per 100 children. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 39 Figure 2.6: Changes in probability of survival to age 5, circa 2010–circa 2020 1.00 Hong Kong SAR, China Azerbaijan Egypt, Arab Rep. Guatemala Probability of Survival to Age 5, circa 2020 Zimbabwe Senegal 0.95 Malawi Burundi Togo Burkina Faso Lesotho 0.90 Chad 0.85 0.80 0.85 0.90 0.95 1.00 Probability of Survival to Age 5, circa 2010 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the probability of survival to age 5, circa 2020 HCI (on the vertical axis), against the probability of survival to age 5, circa 2010 (on the horizontal axis), for 173 countries where child survival data are available for both 2010 and 2020. The dashed line is a 45-degree line; points above (below) represent an increase (decrease) in the probability of survival to age 5 between 2010 and 2020. Blue diamonds in the panels indicate countries for which data are available for both 2010 and 2020, but that are not part of the sample used for the HCI analysis of changes over time because they are missing 2010 comparator data for one of the HCI components. The outlier (blue diamond at far left) is Haiti, where the probability of survival to age 5 was significantly affected by the 2010 earthquake. Malawi, where child survival rates increased from 93 percent in 2014), exclusive breastfeeding (from 91 to 95 percent in the last decade, adopted several 44 percent in 2000 to 70 percent in 2014), preven- evidence-based policies financed by the govern- tion of mother-to-child HIV transmission, and ment and development partners to improve child oral rehydration for diarrhea (up from 48 percent health, including the Accelerated Child Survival in 2000 to 64 percent in 2014), that have in turn and Development Strategy (ACSD), Child Health contributed to improve child survival rates.8 Strategy, Integrated Management of Childhood Illness (IMCI), and the Roadmap to accelerate Fraction of children under 5 not stunted maternal and newborn survival. These policies Advances in health over time are also reflected in and interventions have led to improved cover- decreases in stunting rates for children under 5, age of essential child health services and practices though declines have been modest, on average. across the country, including immunizations (at The fraction of children under 5 not stunted 8 https://www.countdown2015mnch.org/documents/CD_Malawi_July2015_2logos_FINAL2.pdf. 40 Hum an Capital Accumu lati on over T i m e Figure 2.7: Changes in fraction of children under 5 not stunted, circa 2010–circa 2020 1.0 Paraguay North Macedonia Albania Fraction of Children Under 5 Not Stunted, circa 2020 Azerbaijan Congo, Rep. 0.8 Côte d’Ivoire Malaysia Eswatini Zimbabwe Burkina Faso South Africa Indonesia 0.6 Chad Timor−Leste 0.4 0.4 0.6 0.8 1.0 Fraction of Children Under 5 Not Stunted, circa 2010 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the fraction of children under 5 not stunted, circa 2020 HCI (on the vertical axis), against the fraction of children under 5 not stunted, circa 2010 (on the horizontal axis), for 91 countries where the fraction of children under 5 not stunted data are available for both 2010 and 2020. The dashed line is a 45-degree line; points above (below) represent an increase (decrease) in the fraction of children under 5 not stunted between 2010 and 2020. Blue diamonds in the panels indicate countries on which data are available for both 2010 and 2020, but that are not part of the sample used for the analysis of changes over time because they are missing 2010 comparator data for one of the HCI components. is available for comparison between 2010 and from 71 to 62 percent), Papua New Guinea (from 2020 for 91 countries, out of which 42 are in the 56 to 51 percent), Niger (from 56 to 52 percent), 2010–20 HCI comparison sample. Across these Vanuatu (from 74 to 71), Malaysia (from 83 to 79 countries, depicted in Figure 2.7, the fraction of percent), and South Africa (from 75 to 73). children not stunted increased by about 8 per- centage points, on average. The countries with the The overall trend in stunting observed between largest improvements are Côte d’Ivoire (from 61 2010 and 2020 is consistent with its worldwide to 78 percent, an increase of 17 percentage points), decline over the past decades. Progress resulted Sierra Leone (from 56 to 71 percent, a 15 percent- from a variety of factors—not only from over- age point increase), Eswatini (from 60 to 74 per- all economic development but also from health cent, 14 percentage points), and India (from 52 to and nutrition interventions, maternal education 65 percent, a 13 percentage point increase). The and nutrition, maternal and newborn care, and fraction of children not stunted declined in only reductions in fertility or reduced interpregnancy. a small group of countries: Angola (with a decline Given the multiple determinants of stunting, T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 41 multisectoral solutions are necessary. Some exam- from two rural districts in 2011 to 18 rural districts ples are described in Box 2.2. Out of the countries 9 in 2013, eventually reaching 60 districts. The RBF for which stunting data are available, 25 are clas- in Zimbabwe initially focused on reproductive, sified as fragile and conflict affected (FCS). While maternal, newborn, and child health indicators stunting decreased on average in these countries and later expanded to include HIV/AIDS, tuber- too, improvements in FCS countries were in the culosis, malaria and noncommunicable diseases. order of 3.6 percentage points, while in non-frag- The early indications of positive performance ile countries they were in the order of 6.1 percent- under RBF, marked by increased coverage and age points.10 quality of key maternal and child health services (a 13 percentage point increase in institutional Adult survival deliveries in RBF-implementation districts, for Adult survival rates have been improving steadily instance) led to the scale up of RBF implementa- over the last decade. In 2010, 82 percent of tion across the country.11 Maternal mortality also 15-year-olds were expected to survive to age 60, saw declines through the improved coverage of compared with 85 percent in 2020. Figure 2.8 maternal health services facilitated by urban and illustrates the improvement in adult survival rates rural voucher schemes providing care to pregnant over the last 10 years; most countries are above women.12 Another potential contributing factor is the (dashed) 45-degree line. Countries with the the decrease in HIV/AIDS prevalence and reduc- greatest improvements include Eswatini, where tion in HIV/AIDS-related mortality due to the survival rates increased by close to 25 percentage improved coverage of antiretroviral treatment. Of points—although from an extremely low base— all adults aged 15 years and over living with HIV, from 35 to 60 percent, and Zimbabwe, where rates 89 percent were on treatment, while 76 percent of increased from 47 percent to 65 percent. While children aged 0–14 years living with HIV were on most of the countries with large improvements in treatment.13 adult survival are in Sub-Saharan Africa, survival also improved substantially in three countries Eswatini also witnessed some of the largest in Eastern Europe and Central Asia: Kazakhstan improvements in adult survival rates during the (from 76 to 84 percent), Belarus (from 79 to 84 per- decade. However, the country has the lowest adult cent), and Russia (from 75 to 80 percent). survival rate among non-fragile and conflict-af- fected economies in the sample. This reflects the Many factors are behind these trends. In high prevalence of HIV/AIDS, the leading cause Zimbabwe, improvements were fueled by a of deaths in the country.14 Eswatini continues to combination of increased resources allocated to experience the highest rate of HIV/AIDS preva- the health sector along with a progressive focus lence globally, affecting 27 percent of 15-to-49- on results. This included the implementation year-olds.15 The rate of new infections is also the of results-based financing (RBF) approaches in highest in the world, with young women 15-24 health centers and district hospitals, increasing years five times more likely to be infected with 9 Bhutta et al. (2020). 10 This differential persists even when the initial level of stunting and GDP per capita are factored in. 11 World Bank (2016). 12 World Bank (2019). 13 UNAIDS (2020a). https://www.unaids.org/en/regionscountries/countries/zimbabwe 14 CDC (2019). https://www.cdc.gov/globalhealth/countries/eswatini/pdf/eswatini-factsheet.pdf. 15 UNAIDS (2020). https://www.unaids.org/en/regionscountries/countries/swaziland. 42 Hum an Capital Accumu lati on over T i m e Box 2.2: Cross-sectoral interventions to address stunting Through its effects on health and cognitive development, undernutrition early in life stunts children’s development and prevents them from reaching their full potential, in school and during adulthood. According to Bhutta et. al. (2020), interventions that target nutrition both from within and outside the health sector, through improvements in maternal education and nutrition, maternal and newborn care, reductions in fertility or impregnancy intervals, can be effective in reducing stunting in a variety of contexts. The following examples illustrate cross-sectoral engagements to accelerate stunting reduction. Madagascar. With rates as high as 60% in some regions, stunting is a one of the most serious imped- iments to Madagascar’s socio-economic development. The World Bank, with co-financing from The Power of Nutrition, is supporting the Government of Madagascar’s efforts to reduce stunting through the Multiphase Programmatic Approacha (MPA) to Improve Nutrition Outcomes, which aims to reach 75% of children in Madagascar over the next ten years with a high-impact package of services delivered through a strengthened integrated nutrition and health platform. The program evolves based on lessons learned from the field and on scaling-up successful and cost-effective interven- tions. Madagascar’s social safety net programs are also playing an important role in addressing child malnutrition and development. The FIAVOTA safety net program in the drought affected areas of Southern Madagascar had positive impacts on acute malnutrition, while the Human Development Cash Transfer program has had positive impacts on food security as well as young children’s socio-cognitive development, including language learning and social skills. Rwanda. Over the past two decades, Rwanda has registered strong progress on poverty reduction and human development. However, the child stunting rate remains high at 38%, particularly among poorer and larger households. The government has been taking evidence-based action to combat stunting and invest in child development across multiple sectors. Social protection has been central to this effort, striking at the nexus between poverty, vulnerability, and child malnutrition. Rwanda’s flagship social safety net, the Vision 2020 Umurenge program (VUP), has received sustained World Bank support over the years, providing over a million poor and vulnerable people with income sup- port and accompanying measures. In recent years, child- and gender-sensitive safety net interven- tions were introduced in the VUP that are now being expanded. These include Nutrition-Sensitive Direct Support (NSDS) and a Co-responsibility Cash Transfer (CCT), which targets the poorest households with pregnant women and/or children under age two, incentivizing them to access essential health and nutrition services. Rwanda’s game plan also includes strengthening high-im- pact health and nutrition interventions on the supply side, as well as agriculture interventions that improve food security and increase dietary diversity, and pre-primary level education interventions. Pakistan. Fill the Nutrient Gap (FNG), an innovative analysis by the World Food Programme, iden- tifies the bottlenecks that drive malnutrition across the food system, with a special emphasis on the availability, cost, and affordability of a nutritious diet. Using the Cost of the Diet software developed by Save the Children UK, the FNG estimates the minimum cost of a nutritious diet using locally available foods. By comparing this to household food expenditure data, the propor- tion of households unable to afford a nutritious diet is estimated. In Punjab, Pakistan, this exer- cise highlighted that a nutritious diet was unaffordable for two-thirds of the population, with the largest gap for the poorest 20% who are also targeted by the Benazir Income Support Program (BISP). The Government of Pakistan and the WFP jointly evaluated options to complement a cash transfer with nutrition specific interventions, comparing the impact of market-based interventions with a free provision of Specialized Nutritious Foods (SNF), and SNF provision in combination T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 43 with a fresh food voucher. A locally-produced SNF could be an effective way to reduce the nutrient intake gap caused by non-affordability.b For instance, research among pregnant and lactating women and children under-two by Aga Khan University had found impact on some nutritional indicators. Based on this, the Government of Pakistan, together with development partners, designed a nutrition-sensitive conditional cash transfer program targeting pregnant and lactating women (until 6 months after delivery) and children up to 24 months old. The pro- gram included a combination of ante-natal care checkups, immunization, growth monitoring and nutrition education, SNF for women and for children, a small cash transfer to encourage the uptake of the services, and a condition of one-child per household enrolled at a time to encourage birth spacing. The program will be piloted before a nation-wide roll-out. The World Bank will support an impact evaluation to determine cost-effectiveness of interventions. Other initiatives are already ongoing, including a nutrition-sensitive conditional cash transfer pro- grams supported by the World Bank in the Federal territories, Punjab Province and the merged districts of KP province, as well as increasing multisectoral collaboration between the federal government and provincial governments to improve nutrition in Pakistan. a World Bank (2018c) b World Food Programme (2019), and World Food Programme (2017) Figure 2.8: Changes in adult survival rates, circa 2010–circa 2020 1.0 Iran, Islamic Rep. Estonia Kazakhstan Botswana 0.8 Russian Federation Malawi Adult Survival Rate, circa 2020 Congo, Rep. Namibia Cameroon South Africa Zimbabwe Côte d’Ivoire 0.6 Eswatini Lesotho 0.4 0.2 0.2 0.4 0.6 0.8 1.0 Adult Survival Rate, circa 2010 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots adult survival rates circa 2020 HCI (on the vertical axis) against adult survival rates circa 2010 (on the horizontal axis) for 169 countries where adult survival data are available for both 2010 and 2020. The dashed line is a 45-degree line; points above (below) represent an increase (decrease) in adult survival rates between 2010 and 2020. Blue diamonds in the panels indicate countries on which data are available for both 2010 and 2020, but that are not part of the sample used for the HCI analysis of changes over time because they are missing 2010 comparator data for one of the HCI components. 44 Hum an Capital Accumu lati on over T i m e HIV than their male counterparts.16 While the upper-secondary enrollment; among upper-mid- crisis is far from resolved, the country has made dle-income countries, the rise stems from pre-pri- enormous progress in reducing the number of mary and upper-secondary enrollment. AIDS-related deaths, with a 35 percent reduction between 2010 and 2018.17 Economies that have experienced a significant increase in the EYS over the past decade include Adult survival rates declined in only a handful of Bangladesh; Burkina Faso; Côte d’Ivoire; Macao countries, among these Jamaica experienced the SAR, China; and Togo. In Bangladesh, the EYS rose largest decline (less than 1 percentage point). The from 8.2 years in 2010 to 10.2 years in 2020. While United States, where adult mortality rose from 106 many elements are behind this success, the gov- to 110 deaths per 1,000 15-year-olds, is the rich- ernment’s sustained effort to reduce fertility likely est country among this group. In 2020, the adult provided incentives to invest more in children’s survival rate for the United States was significantly schooling. Girls’ participation in secondary school below the level that would have been predicted was also stimulated by the Bangladesh Female based on income.18 Stipend Program, which has enabled the country to achieve one of its Millennium Development Unsurprisingly, child and adult survival improved Goals, gender parity in education.20 together, reflecting a broad improvement in the underlying health status of populations. Of the 103 countries with an HCI in 2010 and 2020, 21 exhibit a lower EYS in 2020 than in 2010. Expected years of school Among these 21 countries, the median coun- Quantity of schooling, as measured by expected try lost 0.09 years of school. Enrollment rates years of school (EYS), increased by about a half year have declined in some richer countries, includ- of schooling (0.47 years to be precise) over the past ing Bulgaria, Luxembourg, Italy, Romania, and decade in the 119 countries for which schooling Ukraine. In Romania in 2010–20, the EYS fell by data are available in 2010 and 2020 (Figure 2.9). 19 0.8 years, largely driven by decreases in primary These gains materialized across all levels of income and upper-secondary enrollment (see Box 2.3). (Figure 2.10). Low-income countries had the larg- est improvement, 0.90 years, mostly due to higher Learning enrollment rates in pre-primary and primary Progress in learning outcomes as measured by har- education. In lower-middle-income countries, monized test scores has been modest over the past the EYS has risen by an average of 0.81 years, and decade. While there are caveats to comparing test most of this increase derives from higher enroll- score over time (see Box 2.4), harmonized test score ment rates in primary and upper-secondary edu- data from comparable testing programs are avail- cation. Upper-middle- and high-income countries, able for 103 countries circa 2010 and circa 2020. which had the highest EYS values at the start of the The average test scores from this sample remained period, experienced the smallest increases since virtually unchanged, at 452 (Figure 1). However, this 2010. Among high-income countries, about 50 per- stable average masks substantial improvements and cent of the rise can be explained by an increase in declines in different countries over the past decade. 16 UNICEF (2020). https://www.unicef.org/eswatini/hivaids. 17 UNAIDS (2020). https://www.unaids.org/en/regionscountries/countries/swaziland. 18 Case and Deaton (2020) connect the decrease in life expectancy in the United States to the “deaths of despair” phenomenon. 19 Refer to appendix C for more details on this calculation and for details on how enrollment data are imputed when missing. 20 See Gribble and Voss (2009) as well as Ubdaidur et al. (1987). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 45 Figure 2.9: Changes in expected years of school, circa 2010–circa 2020 14 Macao SAR, China United Arab Emirates Azerbaijan 12 Egypt, Arab Rep. Expected Years of School, circa 2020 Jordan Timor−Leste 10 Togo Guatemala Cameroon Gambia, The Côte d’Ivoire 8 Burkina Faso 6 4 4 6 8 10 12 14 Expected Years of School, circa 2010 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots expected years of school circa 2020 HCI (on the vertical axis) against expected years of school, circa 2010 (on the horizontal axis) for 119 countries where enrollment data are available for both 2010 and 2020. The dashed line is a 45-degree line; points above (below) represent an increase (decrease) in expected years of school between 2010 and 2020. Blue diamonds in the panels indicate countries on which data are available for both 2010 and 2020, but that are not part of the sample used for the analysis of changes over time because they are missing 2010 comparator data for one of the HCI components. Figure 2.10: Contribution to change in the EYS, by country-income group, 2010–20 1.5 Expected Years of School (years) Pre−primary Primary Lower−secondary Upper−secondary Contribution to change in 1.0 0.5 0.0 Low income Lower-middle Upper-middle High income income income Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: Based on 103 economies with an HCI for 2010 and 2020. Results are the outcome of a Shapley decomposition at the country level and averaged by income group. 46 Hum an Capital Accumu lati on over T i m e Box 2.3: Why have expected years of school decreased in Romania? Three main factors explain why the expected years of school (EYS) in Romania have declined in the past decade (from 12.7 to 11.8 years). First, in the wake of the financial crisis of 2008– 09, a decision was made to close Arts and Crafts Schools, which offered a vocational path as part of upper-secondary education. The number of students enrolled in these schools fell by more than 50 percent between 2010 and 2018, without a corresponding rise in enrollments in other types of upper-secondary education (see Figure 1). While the resident population of school-age children fell by only 7 percent during the decade, net upper-secondary enroll- ment rates fell from 86 percent to 77 percent in 2010–18. In short, the young people who would have enrolled in the vocational schools never enrolled in other schools. In 2015, the three-year vocational path was reintroduced, subsequently helping the system to recover. Second, the number of out-of-school children, including primary-school-age children, has continued to increase during the past decade. Indeed, the number of out-of-school chil- dren ages 6–10 doubled between 2009 and 2018, from 43,000 to 98,000. The underly- ing reasons include persistent underfunding of the sector. Government spending on pre- primary and primary education is the lowest among European Union (EU) countries (see Figure 2). Moreover, Romania still lacks an early warning system to alert authorities about children who are at risk of dropping out. With the help of the World Bank and the European Commission, work is under way to implement such a system.a Third, in 2012, the government introduced a compulsory year of schooling starting at the age of 6 (Romanian National Education Law no 1/2011, article 29, paragraph 2). This meant that, as of 2012, all children age 6 were counted as out of school if they were not in school. In 2018, some parents were still postponing enrolling their children in school at age 6, six years after the implementation of the new law. Figure 1: Dynamics in enrollment Figure 2: Spending on preprimary and numbers in upper secondary primary education (share of GDP) (index, 2010 = 100) 120 5.0 4.5 100 4.0 Index, 2008/09=100 3.5 80 3.0 % of GDP 60 2.5 2.0 40 1.5 1.0 20 0.5 0 0.0 2008/2009 2009/2010 2010/2011 2011/2012 2012/2013 2013a/2014 2014/2015 2015/2016 2016/2017 Romania Bulgaria Lithuania Czechia United Kingdom Germany Finland Greece Malta Ireland France Hungary Austria Slovakia EU average Italy Cyprus Netherlands Portugal Spain Luxembourg Poland Slovenia Belgium Estonia Latvia Croatia Denmark Sweden High school Vocational education Residents age 14–19 2017 2010 Source: INS (National Institute of Statistics, Romania). Source: “General Government Expenditure by Function,” 2018. Statistical Yearbook 2017. Bucharest: INS. [gov_10a_exp], Eurostat, Luxembourg. Source: Contributed by Alina Sava and Lars Sondergaard. a European Commission and World Bank (2018). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 47 Box 2.4: Challenges in test-score comparison over time Using PISA or TIMSS to compare performance of secondary-school students at two points in time may be more complicated in a middle- or low-income country than in a high-income country. In settings where secondary school completion is far from universal, selection bias can affect the results, because assessments like PISA and TIMSS test only enrolled students. The youth who are still enrolled in school at age 15 (PISA) or in grade 8 (TIMSS) are gener- ally those who are from better-off, better-educated households or who have higher ability— which is likely to bias test scores upwards.a This bias causes problems for comparisons not only across countries, but also potentially over time.b If secondary school participation rises significantly between two test rounds, the students who are newly enrolled on the margin will likely score lower on average. This effect will bias downward the change in scores, which would cause PISA or TIMSS to understate the actual system improvement that a con- stant sample of students would have experienced over the same period. For the average country over the past decade, this bias affecting test-score improvements was probably not very large. On average in middle-income countries, lower-secondary completion rates increased only from 76 percent in 2010 to 79 percent in 2018; in low-in- come countries, they rose from 36 percent to 41 percent (World Development Indicators). Nevertheless, the bias could matter over longer periods of time or for countries that have increased secondary-school participation more rapidly. Source: contributed by Halsey Rogers. a Hanushek and Woessmann (2011). b Glewwe, et al. 2017. Of these 103 countries, roughly half (49) saw a drop test scores increasing from 397 (based on PISA in test scores (appearing below the dashed 45-degree 2009) to 434 (based on PISA 2018). Albania’s PISA line in Figure 2.11), while the other half saw small score improvements coincide with the launch of increases. Among countries with improvements in intensive reform efforts in its education sector. test scores, Ecuador’s harmonized test score based The government launched the National Education on the LLECE test went up by 47 points from 373 Strategy (NES) in 2004, which was the first attempt to 420, while Cyprus and Qatar recorded gains of to develop a long-term roadmap for the sector. The about 40 points in harmonized test scores based NES served as a catalyst for a range of reforms that on TIMSS/PIRLS and PISA tests, respectively. continued to be implemented through the Pre- Meanwhile, Egypt and Lebanon saw their harmo- University Education Strategy launched in 2014. nized test scores based on TIMSS/PIRLS decline by These reforms include improved teacher recruit- around 40 points (from 399 to 356 and 428 to 390, ment, compensation, and management; a revised respectively). In Sub-Saharan Africa, test scores in curriculum for basic and general upper-secondary Cameroon, Chad, and Madagascar dropped signifi- education focused on competencies; enhanced cantly between the two rounds of PASEC. transparency and accountability through reform of the Matura (grade 12 exam), the national student Albania witnessed one of the largest improve- assessment; reduced price and improved textbook ments in learning outcomes, with harmonized quality through a reformed procurement process; 48 Hum an Capital Accumu lati on over T i m e Figure 2.11: Changes in harmonized test scores, circa 2010–circa 2020 600 Macao SAR, China Cyprus New Zealand Harmonized Test Scores, circa 2020 500 Seychelles Slovak Republic Greece Albania Ecuador 400 Morocco Cameroon Egypt, Arab Rep. Madagascar South Africa Chad 300 350 400 450 500 550 Harmonized Test Scores, circa 2010 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots harmonized test scores circa 2020 HCI (on the vertical axis) against harmonized test scores circa 2010 (on the horizontal axis) for 103 countries where harmonized test score data are available for both 2010 and 2020. The dashed line is a 45-degree line; points above (below) represent an increase (decrease) in harmonized test scores between 2010 and 2020. Blue diamonds indicate countries on which data are available for both 2010 and 2020, but that are not part of the sample used for the HCI analysis of changes over time because they are missing 2010 comparator data for one of the HCI components. provision of textbook subsidies to the poorest However, changes in learning and in years of households; a stronger focus on inclusive educa- education appear to be positively correlated tion; and expansion of enrollment in pre-primary in upper-middle- and high-income countries and upper-secondary education. 21 and (albeit very weakly) negatively correlated in lower-middle- and lower-income countries.22 A question that is often part of policy discus- While this evidence is suggestive at best, it points sions is whether improvements in school access to the need to understand more clearly how edu- are associated with drops in learning. In this cation systems can be strengthened in poorer sample, there is no clear correlation between countries to achieve high-quality learning at the changes in years of education and test scores. same time that access is being expanded. 21 Ministry of Education and Science, Republic of Albania (2005). https://planipolis.iiep.unesco.org/sites/planipolis/files/ressources/ albania-education-strategy-2004-2015.pdf. 22 Test scores and years of schooling series are negatively correlated within Latin America and the Caribbean (correlation of −0.16), the Middle East and North Africa (−0.28), and Sub−Saharan Africa (−0.14). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 49 Figure 2.12: Changes in HCI components and income, 2010–20 a. Probability of Survival to Age 5 b. Expected Years of School 1.00 14 HIC HIC Probability of Survival to Age 5 0.98 UMIC 12 Expected Years of School UMIC 0.96 10 LMIC LMIC 0.94 8 0.92 LIC 0.90 LIC 6 6 8 10 12 6 8 10 12 Log GDP Per Capita Log GDP Per Capita c. Harmonized Test Scores d. Fraction of Children Under 5 Not Stunted 1.0 600 Fraction of Children Under 5 Not Stunted 550 Harmonized Test Scores HIC 500 UMIC 0.8 450 UMIC LMIC 400 LIC LMIC 350 300 0.6 LIC 6 8 10 12 6 8 10 12 Log GDP Per Capita Log GDP Per Capita e. Adult Survival Rate 1.0 HIC Adult Survival Rate UMIC 0.8 LMIC LIC 0.6 6 8 10 12 Log GDP Per Capita Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: Each panel plots the component average for income groups using the World Bank Group classification (on the vertical axis) against log real GDP per capita (on the horizontal axis) for countries where data are available for both 2010 and 2020. The 2010 HCI is denoted by dots and the HCI 2020 is denoted by an arrow. Panel a calculates income-group averages for the probability of survival to age 5 for 173 countries where data were available. Panel b calculates income-group averages for expected years of school for 119 countries where data were available. Panel c calculates income-group averages for harmonized test scores for 103 countries where data were available. Panel d calculates income-group averages for the fraction of children under 5 not stunted for 91 countries where data were available. Panel e calculates income-group averages for adult survival rates for 169 countries where data were available. LIC = low-income countries; LMIC = lower-middle- income countries; UMIC = upper-middle-income countries; HIC = high-income countries. 50 Hum an Capital Accumu lati on over T i m e 2.2.3 Dimensions of human capital and to relatively small increases in GDP. It stayed vir- economic development tually unchanged for middle- and high-income Much like the overall HCI, changes in individ- countries. ual measures of human capital don’t happen in a vacuum and are correlated with changes in Reconstructing this picture at the country level income. Using a similar visualization as in the in Figure 2.13 reveals significant heterogeneity, previous chapter, Figure 2.12 illustrates the aver- including dramatic improvements in outcomes age improvements in the Index components as despite little improvement in income (this is the per capita income rises. For example, in panel case, for example, of survival in Eswatini). No a, child survival rates are plotted against log real country, however, showed large GDP improve- GDP per capita. A line connects the solid dots indi- ment without at least some improvement in some cating the group average in 2010 to the arrows human capital dimension. indicating the average in 2020. The lines all slope upward, reflecting the pattern of improved child 2.2.4 Socioeconomic differences and survival globally. The lines also become shorter as progress in human capital they approach the top of the panel where there is Regional and national averages provide important less room for improvement. The gradient of the insights into development trajectories over time. lines is also of interest, reflecting the rate at which However, they also mask the differential trends in outcomes improved with changes in per capita human capital across groups within countries, par- GDP. The steep lines, such as those for low-in- ticularly between richer and poorer households. come countries and Sub-Saharan Africa, showcase The HCI relies on component data from admin- large increases in child survival rates despite rela- istrative sources that cannot readily be disaggre- tively small gains in per capita GDP. This is likely gated by socioeconomic status. Survey data—par- a reflection of improvements in global health, ticularly from Demographic and Health Surveys such as better but less expensive technologies.23 and Multiple Indicator Cluster Surveys—also Conversely, flatter slopes in high-income coun- measure child survival rates, enrollment rates, and tries, Europe and Central Asia, and North America stunting rates disaggregated by quintiles of socio- suggest smaller gains in the outcome relative to economic status. While these survey estimates are increases in per capita GDP. The arrows are also not always directly comparable with administra- shorter, because these countries were already near tive data, they can provide insights into the rates full child survival in 2010. of change in outcomes for the richest and poorest households within countries and how these affect The patterns are similar (upward sloping with national averages. decreasing slopes as income increases) for adult survival and the absence of stunting across these This subsection discusses child survival, enroll- income groups, though adult survival and child ment, and fraction of children not stunted dis- survival rates share the feature of steeper improve- aggregated by socioeconomic status, based on ments at low income levels. Learning in low-in- Demographic and Health Surveys and Multiple come countries dropped marginally with respect Indicator Cluster Surveys for selected countries 23 For the role of technology in the progress in child survival, see Jamison et al. (2016). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 51 Figure 2.13: Changes in HCI components and income, circa 2010 and circa 2020, individual trajectories a. Probability of Survival to Age 5 Luxembourg b. Expected Years of School 1.00 14 Lithuania Ukraine Solomon Islands Hong Kong SAR, China Probability of Survival to Age 5 Azerbaijan 12 Namibia 0.95 Expected Years of School Zimbabwe Iran, Islamic Rep. Saudi Arabia Gabon Lao PDR Macao SAR, China Eswatini 10 Togo Malawi 0.90 Lesotho Namibia Equatorial Guinea Niger 8 Mali Nigeria Botswana Togo 0.85 Chad Senegal Burundi 6 Côte d’Ivoire 0.80 Chad Haiti 4 6 8 10 12 6 8 10 12 Log GDP Per Capita Log GDP Per Capita c. Harmonized Test Scores d. Fraction of Children Under 5 Not Stunted 600 1.0 Fraction of Children Under 5 Not Stunted Korea, Rep. Singapore Turkey Harmonized Test Scores Estonia Malaysia 500 Lithuania 0.8 Libya Luxembourg Gambia, The Turkey Ecuador Bolivia Iran, Islamic Rep. Senegal Eswatini Uganda 400 Indonesia 0.6 Central African Republic Indonesia Eswatini Kuwait Papua New Guinea Uganda Ecuador Malawi India Malawi Gambia, The Burundi Yemen, Rep. 300 0.4 6 8 10 12 6 8 10 12 Log GDP Per Capita Log GDP Per Capita e. Adult Survival Rate 1.0 China Lebanon United States 0.8 Yemen, Rep. Belarus Afghanistan Russian Federation Adult Survival Rate Burundi Equatorial Guinea Botswana 0.6 Malawi Namibia Zimbabwe 0.4 Lesotho Eswatini 0.2 6 8 10 12 Log GDP Per Capita Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: Each panel plots the country-level averages for each component (on the vertical axis) against log real GDP per capita (on the horizontal axis) for countries where data are available for both 2010 and 2020. The 2010 HCI is denoted by dots and the HCI 2020 is denoted by an arrow. Panel a shows the probability of survival to age 5 for 173 countries where data were available. Panel b shows expected years of school for 119 countries where data were available. Panel c shows harmonized test scores for 103 countries where data were available. Panel dshows the fraction of children under 5 not stunted for 91 countries where data were available. Panel e shows adult survival rates for 169 countries where data were available. 52 Hum an Capital Accumu lati on over T i m e Figure 2.14: HCI components disaggregated by socioeconomic status 1.00 a. SES-Disaggregated Child Survival Rate b. SES-Disaggregated Expected Years of School 12 2006 2000 0.95 2014 10 Expected Years of School 2015 2014 SES−Disaggregated 2015 2012 2007 SES−Disaggregated Child Survival Rates 2012 2012 2010 2013 2004 0.90 2005 8 2010 2005 2006 2000 6 0.85 2004 2010 4 0.80 2000 2003 2 0.75 6 8 10 6 8 10 Log Real GDP Per Capita at PPP Log Real GDP Per Capita at PPP Malawi Haiti Senegal Burkina Faso Bangladesh Azerbaijan c. SES-Disaggregated Fraction of Children Not Stunted 0.9 Fraction of Children Not Stunted 2016 2014 0.8 2011 SES−Disaggregated 2011 2016 0.7 2005 2011 2006 0.6 2000 0.5 6 8 10 Log Real GDP Per Capita at PPP Uganda Côte d’Ivoire Congo, Rep. Source: World Bank calculations based on DHS/MICS data as reported in Wagstaff et al. 2019 (for child survival rates and fraction of children under 5 not stunted) and Filmer and Pritchett (1999) and subsequent updates (for expected years of school). Notes: The figure plots selected HCI components disaggregated by quintile of socioeconomic status (vertical axis) against log real GDP per capita (horizontal axis). The solid dot indicates the average across quintiles, and the top (bottom) end of the vertical bar indicates the value for the top (bottom) quintile. Colored bars show the spread of components over time. Panel a shows SES-disaggregated child survival rates for Haiti, Malawi, and Senegal. Panel b shows SES-disaggregated expected years of school for Azerbaijan, Bangladesh, and Burkina Faso. Panel c shows SES-disaggregated fraction of children under 5 not stunted for CÔte d’Ivoire, the Republic of Congo, and Uganda. with large changes in outcomes in the HCI dataset.24 most of the examples come from these coun- Because these surveys are fielded predominantly tries. Figure 2.14 reports human capital outcomes in low- and lower-middle-income countries, over time, disaggregated by socioeconomic status 24 School enrollment data by age disaggregated by socioeconomic status come from the latest update to the household wealth and education- al attainment dataset first described in Filmer and Pritchett (1998). The latest version of their dataset contains 345 Demographic and Health Surveys and Multiple Indicator Cluster Surveys (DHS/MICS), with enrollment rates for 99 countries over 1990–2017. The child (under-5) mortality rates and stunting rates disaggregated by socioeconomic status come from the latest edition of the HEFPI database described in Wagstaff et al. (2019). Both datasets calculate the socioeconomic status index in the same way, using principal component analysis to aggre- gate responses to questions on asset ownership and housing characteristics into a household-level socioeconomic status index. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 53 Box 2.5: Transforming a low-performing education system into Brazil’s best school network Ceará is a northeastern state in Brazil that improved its education outcomes much faster than the rest of Brazil, in just over a decade. Home to 9 million people (4 percent of the population of Brazil) and with the fifth-lowest GDP per capita in the country, almost all of Ceará’s 184 municipalities had low levels of quality in teaching and very limited resources, spending about one-third less in per-student education than wealthier Brazilian states such as São Paulo. Among these municipalities is Sobral, 200,000 inhabitants, which in the late 1990s suffered from a highly fragmented school system, with many poorly maintained small schools, most of which were in rural areas and had multi-grade classes. Despite a reorganization of the school network, a 2005 diagnostic found that 40 percent of grade 3 children were not able to read, 32 and 74 percent of students in primary and lower-secondary schools, respectively, were over grade-appropriate ages, and 21 percent of lower-secondary school students dropped out. Between 2005 and 2015, Sobral managed to achieve remarkable progress in educational outcomes. In 2005, Sobral ranked 1,366th in education among Brazilian munici­ palities.a A decade later, it ranked first among 5,570 municipalities in the country in both primary and lower-secondary education, achieving learning outcomes comparable to world- class education systems as measured by PISA. Today, although its per capita GDP amounts to little over half the national average,b Ceará has the lowest rate of learning poverty in Brazil, and Sobral has some of the country’s best primary schools. Education outcomes in both the region and the municipality exceed all expectations, given the socioeconomic con- text in which students live and learn: Sobral’s student-to-teacher ratio is relatively high, at 28.9, compared with 21.0 in Ceará and 20.3 on average in Brazil as a whole. These points suggest a high efficiency of the education system. Ceará’s approach was driven by a mix of the following elements, whose effectiveness is sup- ported by international evidence: 1. The provision of fiscal and nonmonetary incentives for municipalities to achieve education outcomes; 2. Technical assistance to municipal school networks to enhance teacher effectiveness and achieve age-appropriate learning; 3. The regular use of a robust monitoring and evaluation system, followed by adequate action; and 4. Giving municipalities autonomy and accountability to achieve learning:c in Ceará, unlike the rest of Brazil, municipalities are responsible for the entirety of the education provided, from pre-primary to lower-secondary school. A key factor enabling Ceará to emerge as one of Brazil’s top performers in education has been the capacity of state political leaders to insulate education from partisan politics. This has contributed to strong, sustained political leadership committed to improving the quality of education. Sobral organized its education policy under four pillars: 1. Continuous use of student assessments; 2. A focused curriculum with a clear learning sequence and prioriti- zation of foundational skills; 3. A pool of well-prepared and motivated teachers; 4. A sys- tem of autonomous and accountable school management with school principals appointed through a meritocratic technical selection process.d The municipality’s goal was to achieve the universal completion of lower-secondary education at the right age with appropriate learning. The results obtained show the effectiveness of goal setting and the importance of political leadership for education outcomes. 54 Hum an Capital Accumu lati on over T i m e The COVID-19 pandemic threatens the progress made by Ceará. A recent study shows that two to three weeks of school closures in São Paulo amid the H1N1 pandemic resulted in an estimated two months in learning loss. Using this as a proxy for the COVID-19 pandemic, the paper concludes that an estimated two to three months school closure could induce a learning loss equivalent to a half-semester of a school year in Brazil (World Bank, 2020a). However, Ceará’s progress and the pillars that led it there should help the region tackle the tough job that lies ahead once the pandemic subsides. Source: Based on Cruz and Loureiro (2020), and World Bank (2020). a As per Brazil’s Basic Education Development Index, IDEB. b Ceará’s per capita GDP was USD PPP 8,068 in 2019, compared with USD PPP 10,666 in Sobral and USD PPP 15,662 in Brazil. c https://blogs.worldbank.org/education/there-no-magic-formula-brazils-ceara-and-sobral-success-reduce-learning-po verty?token=53176c4095d917916aa31ea735b5ceaa. d ibid. against log GDP per capita. Panel a shows child improved, the size of the gap between the rich and survival rates, panel b shows EYS, and panel c the poor remained constant.25 fraction of children under 5 not stunted. Each panel shows the country averages over time as There is similar variation in trends in the EYS.26 dots. The top horizontal line reports the outcome Burkina Faso was able to raise the EYS by two for the richest quintile, while the bottom horizon- years, but the gap between rich and poor house- tal line reflects rates in the poorest quintile. holds was maintained at six years. In contrast, Bangladesh was able to increase the average EYS In the case of child survival, Haiti made massive and also cut the gap between the richest and poor- strides between 2000 and 2012, increasing sur- est households in half, from four to two years, vival rates from 86 to 91 percent. Between 2000 between 2004 and 2016. Azerbaijan improved the and 2015, survival rates in Malawi rose from 80 EYS by one year, but the gap between rich and to 93 percent. In Senegal, rates increased from poor households rose from 0.5 years to 1.0 year. 87 to 94 percent between 2005 and 2015. However, Box 2.6 offers an example from Sierra Leone of while each country showed declines in child mor- how a well-designed intervention can contrib- tality, the composition of these changes was quite ute to improve education outcomes for the most different. In Malawi and Senegal, the length of disadvantaged. the bars, that is, the gap between rich and poor households, shortened over time because the The fraction of children under 5 not stunted also increase in the average child survival rate was increased in most countries in the last decade, as in driven by improvements in outcomes among the the cases of Côte d’Ivoire, the Republic of Congo, poorest households. In Haiti, while average rates and Uganda. In Côte d’Ivoire, the average fraction of 25 However, the increase in child mortality in 2010 in the aftermath of the earthquake in Haiti was massive. 26 The EYS data used to calculate the HCI rely on administrative data on pre-primary through upper-secondary enrollment, covering the 4–17 age range for a maximum of 14 years of school. By contrast, DHS/MICS surveys collected enrollment data for children aged 6 to 17 for a maximum of 12 years of school. As a result, the EYS reported in the HCI, calculated using administrative data, cannot be compared to the EYS reported in this section, computed using survey data. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 55 Box 2.6: The immediate effects of providing free education in Sierra Leone Although the majority of children in Sierra Leone start school, few successfully complete their secondary school education, and learning outcomes are among the lowest in the world, contributing to a significant human capital gap. The most cited reason why children drop out of school is not poor quality, however, but cost. Although out-of-pocket expen- ditures on education are fairly low, both in absolute terms and as a percent of household expenditure (about 3% across income groups), they can still represent a significant barriers for poor families, especially given that school fees are due in September, at the height of the hungry season. The flagship program of the government is the Free Quality School Education Program, which was launched in September 2018. It provides selected public schools with block grants (calculated on a per-pupil basis) and school materials, such as textbooks, while mandating that these schools not charge fees. The program seeks to reduce out-of-pocket household spending on education (the “free” component in the program’s name) by eliminating or at least reducing school fees. It also seeks to raise the quality of education (the “quality” com- ponent), through the provision of textbooks and other measures. Public messaging around the program has stressed boosting enrollment: there is now no reason for parents not to send their children to school. Earlier studies of free education in Sub-Saharan Africa looked at cohorts and focused on the long-term impacts following several years of implementation. Data collected in February and March 2019 allows us to assess the effects of free schooling on out-of-pocket house- hold expenditures and enrollment in the first term of the program (beginning Sept 2018), as over 4,000 households that had been interviewed for the 2018 Sierra Leone Integrated Household Survey were re-interviewed then. For each child, the specific school they attended for the 2017/2018 and 2018/19 school years was recorded and linked to the Annual School Census to determine whether the school benefited from the FQSE program in the first term of 2018/19. The main impact of the Free Quality School Education Program in the first term appears to have been a substantial reduction in out-of-pocket education expenditures by households. Over 90 percent of students at public primary and secondary schools receiving the block grants report that they are not paying school fees, up from about a third of primary-school students and almost no secondary-school students in the school year prior. In addition, about two-thirds of students at public schools not yet supported under the program also report that they do not pay school fees. The financial benefits of the program, in terms of reduction in out-of-pocket expenditures, are shared fairly evenly across the welfare distri- bution, although the poorest 20 percent of households receive the largest benefit as a per- centage of total consumption. Administrative data show a large increase in the number of students, but data collected from households reveals no significant change in net or gross enrollment rates. This discrepancy is not unexpected: a young and growing population like Sierra Leone’s will naturally see an increase in the number of school age students each year, and the way the program is struc- tured gives schools an incentive to maximize their reported enrollment. In any case, there was little room for an increase in enrollment, as these rates were already high, particularly for 56 Hum an Capital Accumu lati on over T i m e primary schools. Increases in secondary school enrollment can only come over time as more students successfully reach this level. There has been a small rise in the percent of 5- to 7-year- olds who start school for the first time; this is concentrated among the poorest households. While the FQSE project has reduced out-of-pocket expenditures, the most keenly felt bar- rier to education for households, it remains to be seen whether this will eventually result in higher enrollment rates at the secondary level and higher levels of secondary school com- pletion, and whether the program will be successful in improving the quality of education these students receive. Source: Contributed by Alejandro de la Fuente based on de la Fuente, Foster, and Jacoby (2019). children not stunted increased from 72 percent to 80 countries, government redistributive policies do, percent between 2011 and 2016, but the 25 percent- on average, as good a job of reducing human cap- age point gap between rates in rich and poor house- ital inequality as does increased national income.27 holds remained unchanged. By contrast, Uganda was At the same time, the experiences of countries like able to increase the fraction of children not stunted Senegal, Bangladesh, and Uganda show that coun- while also modestly closing the rich-poor gap. The tries can sometimes decouple children’s human gap narrowed from a difference of 20 to 16 percent- capital outcomes from the income differences age points between 2000 and 2016. In the Republic among their households. of Congo, the rich-poor gap initially increased, as the fraction of children not stunted increased from The following section takes an in-depth look at 71 percent to 78 percent between 2005 and 2011. the experiences of a selected set of countries to However, the country was able to maintain momen- understand how concerted government action can tum in reducing stunting while also reducing the dif- deliver marked improvements in national out- ference between rich and poor households from 24 comes linked to human capital over time and also to 16 percentage points between 2011 and 2014. reduce rich-poor gaps within countries to achieve greater equity. This analysis highlights that countries vary signifi- cantly in the extent to which gains in human capi- tal outcomes are distributed across the population. 2.3 A LONGER-RUN VIEW OF COUNTRY Addressing these rich-poor gaps in human capital PROGRESS28 must remain a priority for governments because, in many cases, the returns to investment in the Human capital is a central driver of sustainable human capital of disadvantaged groups, especially growth and poverty reduction. However, even for early in life, are the highest. However, related evi- governments that recognize the importance of dence shows that, among low- and middle-income investing in the human capital of their citizens, 27 D’Souza, Gatti, and Kraay (2019). 28 The analysis in this section is based on four country case studies produced as part of a series titled Building Human Capital: Lessons from Country Experiences. These are “How Singapore Does It” by Shahid Yusuf, “The Trajectory of Human Capital Development in the Philippines” by Elizabeth M. King, “Morocco: Achievements and Challenges” by Mohamed Benkassmi and Touhami Ab- delkhalek, and “Ghana’s Recent Human Capital Improvements” by Niels-Hugo Blunch. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 57 the process of designing policy and building insti- investments in key human capital outcomes in tutions that foster human capital accumulation recent years. However, they vary considerably in can be complex, with the benefits taking years and their levels of development, their choice of poli- even decades to materialize. This is evidenced in cies and programs to develop human capital, and the earlier sections of the report that show only the outcomes they achieved. modest annual progress for the average country on the HCI. With a score of 0.88, the Southeast Asian island state of Singapore is one of the top performers on A comprehensive understanding of how coun- the HCI. It has a population of 5.7 million and a tries can improve their human capital outcomes per capita GDP at 2011 PPP of US$96,477,30 mak- requires an analysis that adopts a longer time ing it the richest of the four countries studied here. frame and identifies the many aspects of gov- Singapore has built a world-class education system ernment intervention that can lead to positive with an increasing emphasis on analytical skills, change. By allowing a richer understanding of teamwork, and creativity. The success of these countries’ development trajectories, identifying efforts is evident in the increase of mean years of the policies and institutions that proved critical schooling from 4.7 in 1980 to over 11.2 in 2019.31 to improving outcomes, and documenting the In the health sector, Singapore’s life expectancy at challenges involved in maintaining momentum, a birth increased from 67 in 1965 to 83 in 2017, while comparative case study approach offers this depth infant mortality has been on a downward slope, of information. from 27 in 1965 to 2 in 2017.32 Despite this enviable position, the country’s prime minister has stated This section presents the experiences of four coun- that “the job is never done,”33 identifying active tries that have made notable improvements in healthy aging and early childhood education as their key human capital indicators over roughly the areas for improvement. last decade: Singapore, the Philippines, Morocco, and Ghana. The case studies illustrate how policies, The Philippines, with a population of 104.9 mil- programs, and processes that the governments of lion, is the eighth most populous country in Asia these countries adopted improved human capital (and the most populous country included in this outcomes, documenting three interrelated aspects analysis) and has a per capita GDP at 2011 PPP of of the countries’ trajectories: Continuity—sustain- US$8,123. The country’s HCI score of 0.52 means ing effort over many political cycles; coordina- that children born in the country today will fail to tio—ensuring that programs and agencies work achieve almost half their potential. The impor- together; and evidence—building an evidence base tance that governments in the 1970s accorded to to improve and update human capital strategies. 29 mass education in the country jump-started an expansion in school enrollment, with primary The four countries featured in this section gross enrollment rates at about 100 percent and were selected because they have all prioritized rates nearing 90 percent at the secondary level in 29 This approach is based on that used in the World Bank’s Human Capital Project (HCP), taking a whole-of-government approach. 30 World Bank national accounts data, and OECD National Accounts data files. 31 Statistics Singapore https://www.tablebuilder.singstat.gov.sg/publicfacing/displayChart.action. 32 World Development Indicators. 33 https://www.straitstimes.com/singapore/world-bank-ranks-singapore-tops-in-human-capital-index. 58 Hum an Capital Accumu lati on over T i m e 2017.34 However, while access has increased, qual- 5 has fallen significantly, from 22.7 percent in 2011 ity remains an issue, with 15-year-old Philippine to 17.5 percent in 2017.37 students scoring lower than students in nearly all other participating countries in the latest round of The trajectories of policies in these countries indi- PISA in 2018. cate a strong focus on continuity of government sup- port across political cycles, coordination between Morocco, located in the Maghreb region of Africa, sectoral programs and among different levels and has a population of 35.7 million and a per capita branches of government, and evidence-based pol- GDP at 2011 PPP of US$7,641. The country’s com- icies. While all four countries did not implement mitment to human capital development has led to all of these policy directions, the case studies point remarkable gains in the health of its citizens. The to the whole-of-government as an approach with government has launched efforts to combat child enormous potential to build human capital in a and maternal mortality while controlling fertility wide variety of development contexts. rates through intensive, sustained family plan- ning programs. A diligent immunization policy Sustaining political commitment to human has meant that 91 percent of Moroccan children capital development are now fully immunized.35 These efforts have Continuity of commitment and effort over suc- improved human capital outcomes for the coun- cessive governments is key to reaching any long- try, reflected in an HCI score that increased from term goals, but especially in growing human cap- 0.45 in 2010 to 0.50 in 2020. 36 ital, which can take decades and even generations. While not all politically stable countries were able Finally, Ghana in West Africa has a population to maintain a sustained focus on human capital, of 28.8 million and a per capita GDP at 2011 PPP ensuring this continuity is easier if the country in of US$5,194, making it the country with the low- question enjoys political stability, as in the cases of est income in this sample. Despite limited fis- Singapore and Ghana, the latter characterized by a cal space, Ghana’s commitment to improving stable, multiparty democracy since 1992. human capital and innovative policies have led to marked improvements in the outcomes of its citi- By contrast, a consistent approach to building human zens. Since the government introduced education capital has been harder to achieve in Morocco, where reforms after a major national economic crisis in political commitment to education across successive 1983, primary enrollment rates have increased governments did not extend to other policies crit- substantially, for example, from 67 percent to ical to improving human capital outcomes. In the 95 percent between 2000 and 2017. Increasing Philippines, although several successive political school enrollments and increased access to edu- administrations have adopted and sustained robust cation have led to an influx of students who are strategies to build the human capital of the popula- more likely to come from disadvantaged families. tion, they have not succeeded in growing sufficiently Despite this, Ghana’s harmonized test scores have the capacity and good governance needed to imple- not declined. Stunting in children under the age of ment these efforts on the ground. 34 World Bank data: https://data.worldbank.org/indicator/SH.STA.STNT.ZS?locations=PH-1W-Z4; https://data.worldbank.org/indi- cator/SE.PRM.NENR?locations=PH-1W-Z4. 35 Enquête Nationale sur la Population et la Santé Familiale (ENPSF, 2011 et 2018). 36 Note however, that as indicated in appendix C, as in a handful of other countries, comparisons for learning in Morocco refer to different international testing programs. 37 UNICEF/WHO/World Bank Joint Child Malnutrition Estimates ( JME). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 59 In addition to political commitment, human cap- has set a high standard in this respect by ensuring ital development requires adequate and sustain- that expenditures are tightly managed, including able funding. In particular, domestic resources by imposing severe sanctions for corrupt practices. are central to achieving development objectives. Countries can enhance the quality and foster While successive governments in the Philippines the legitimacy of tax systems by strengthening have enacted human capital development laws the operational capacity of tax administrations. that reflect principles similar to those espoused However, doing so can be a challenge for develop- by more successful countries, they have gener- ing countries with limited resources. Some coun- ally failed to provide adequate financing to ensure tries have also found innovative ways to finance effective implementation. The country spends the necessary policies.38 4.4 percent of its GDP on its health programs and 3.5 percent on education programs, compared For example, the Singapore government has been with an average of 6.5 percent and 4.5 percent, able to mobilize domestic resources through the respectively, for an average country at the same Central Provident Fund (CPF), which has played income level. This has resulted in understaffed a critical role in financing infrastructure, hous- and overcrowded clinics and schools, underpaid ing, and other vital investments. Each individual providers, inadequate infrastructure, and a lack of and his or her employer make monthly contri- administrative and technical capacity, especially butions to the CPF that are distributed among at local schools and health facilities. The absence three accounts owned by the individual: (i) an of adequate funding has also hampered efforts to ordinary account for housing and retirement improve governance. Widespread fraud in the dis- purposes; (ii) a special account that is primarily tribution of textbooks, theft of funds or supplies, for retirement; and (iii) a Medisave account that and ghost workers (workers who are paid but do is used to cover medical expenses. The govern- not carry out their jobs) in municipal health facil- ment supplements the contributions of low-in- ities are all reflected in the country’s outcomes. In come earners through a workfare scheme and the PISA 2018 exam, about four-fifths of students adds to Medisave savings of senior citizens. The (81 percent) achieved lower than a minimum level CPF has also underpinned health care financ- of proficiency in reading, while a similarly high ing through Medisave and has fostered citizens’ percentage of students performed below the min- responsibility for their own welfare. Thus, pol- imum level of proficiency in mathematics. icy makers have managed to contain the cost of providing the country’s entire population with The lack of adequate financing—resulting in affordable, high-quality primary health care understaffed facilities, underpaid providers, and by tailoring subsidies to the patient’s age and overcrowded clinics and schools—has particularly ability to pay and charging users high copay- affected the country’s low-income households and ments financed from mandatory health savings more remote regions, which now lag behind the accounts. Regulation and bulk buying of drugs rest of the country in terms of access to services. have also kept pharmacy costs in check. By contrast, Ghana’s innovative funding mech- anism—the National Health Insurance Scheme Levels of funding are crucial, but so is using (NHIS)—was designed to expand primary care resources efficiently. The government of Singapore coverage while also reducing inequity in access to 38 Junquera-Varela et al. (2017). 60 Hum an Capital Accumu lati on over T i m e health care by exempting the poor from premi- to assess the strengths and weaknesses of their ums. The NHIS is funded mainly by a 2.5 percent 39 own institution and to track student performance VAT on selected goods and services, 2.5 percent (using a Pupil Data Bank). The system has enabled from the Social Security and National Insurance the ministry to keep closer tabs on how individual Trust (SSNIT) (largely paid by formal sector work- schools are faring. ers), and the payment of premiums. These funds enable the NHIS to provide prenatal and postna- In Ghana, the government used data to effectively tal care, maternal health care, vaccinations, and retarget school feeding efforts under the Ghana health and nutrition education, all of which may School Feeding Program (GSFP) after it found that have helped to reduce stunting rates in Ghana. As a the targeted population (the poor) was not being result of the NHIS, the government has been able reached. Data from national poverty statistics and to devote a high percentage of its spending to the a food security and vulnerability analysis were health budget (10.6 percent as of 2013), which has combined to refine targeting and reduce leakag- helped to bring down the rate of childhood stunt- es.40 After the retargeting exercise was completed, ing in Ghana in both absolute and relative terms. as of 2013, about 70 to 80 percent of the GFSP was being received by the poorest communities.41 In Collecting and using evidence to inform policy Morocco, on the other hand, a paucity of data has making stymied improvements to the country’s Tayssir Collecting data to inform policy implementation conditional cash transfer program. The Audit and design is easier in a compact city-state like Office of Morocco (as cited in Benkassmi, 2020) Singapore than in a sprawling island nation like explicitly stated in its 2016-17 report that “no the Philippines, but digital technologies are mak- quantifiable indicators are available to monitor ing it easier for all countries to collect and analyze the different programs and prepare annual prog- data and to use the resulting evidence when mak- ress and financial reports that enable evaluation of ing policies and decisions. the performance of these programs.”42 Singapore’s public agencies and statutory boards, “Whole of government” approaches: Adopting its state-of-the-art digital technology, tech-savvy coordinated, multisectoral strategies administrators, and experienced teachers form Multisectoral strategies are most likely to effectively a robust data-collection infrastructure that feeds address the complex underlying determinants of critical information to policy makers in real time. human capital outcomes. Policies that cut across Policy makers use these data to assess school and sectors and lines of authority can also be especially student performance, control costs, help managers beneficial to countries such as the Philippines that and teachers to make decisions at every level, and have limited resources and technical and admin- do workforce planning. For example, the Ministry istrative capacity. In the last 40 years, successive of Education has installed an information-gather- governments in the Philippines have adopted poli- ing mechanism that helps school administrators cies that involved more than one sector, promoted 39 Not everybody has to pay the NHIS premium. Pregnant women are exempt, as are people under 18 years of age and people age 70 and above, and individuals who are employed in the formal sector and contribute to the SSNIT. Additionally, individuals con- sidered too poor to pay are also exempt from paying the premium. This includes beneficiaries of the Livelihood Empowerment Against Poverty (LEAP) program. 40 WFP (2013) and Drake et al. (2016). 41 WFP (2013). 42 Audit Office of Morocco (2017) as cited in Benkassmi (2020), “Building Human Capital - Lessons from Country Experiences: Mo- rocco”. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 61 integrated approaches, and encouraged greater food security, community participation, and participation by stakeholders in service delivery. women’s empowerment. Specifically, it has helped In addition, many policies reflect the fact that fac- reduce short-term poverty and food poverty at tors beyond the social sectors affect human capital the national scale by up to 1.4 percentage points development, such as clean air, a safe water supply, each—a substantial reduction, given that pre-Pan- and the provision of sanitation services. tawid rates were at 26.4 percent for total poverty and 12.5 percent for food poverty.45 The country has several programs that are organized on multisectoral lines. An example is the Pantawid Ghana’s progress in decreasing stunting rates46 has Pamilya Pilipino Program (4Ps), which provides cash also been due in large part to the multisectoral to chronically poor households with children aged approach taken by policy makers. For example, between 0 and 14 years old who live in poor areas. 43 the Ghana School Feeding Program (GSFP) links In return, the beneficiary households are required school feeding programs with agriculture devel- to undertake certain activities aimed at improving opment, especially smallholder production, thus their children’s health and education, such as taking helping to create new markets for locally grown them to health centers regularly, sending them to food.47 Thus, the GSFP spans three different sec- school, and going to prenatal checkups in the case of tors—agriculture, education, and health.48 Also, pregnant women. Thus, 4Ps integrates human capi- initiatives aimed at improving water sanitation and tal development with poverty reduction efforts. The hygiene in schools have helped to increase access Department of Social Welfare and Development to water and sanitation, which is a proven factor in (DSWD) was charged with leading the program’s improving health and education indicators. implementation, and worked with the Department of Health, Department of Education, Department of The experiences of the four countries examined the Interior and Local Government, and the govern- here highlight the importance of sustained effort to ment-owned Land Bank of the Philippines. The 4Ps improve human capital outcomes across political also actively involved local service providers (such cycles, sufficient resource mobilization and effective as school principals and midwives) in implementa- allocation across programs, data and measurement tion by tasking them with verifying that households to inform and design, and multisectoral strategies were fully complying with the prerequisite condi- that address the complex underlying determinants tions for the cash transfers. 44 of human capital outcomes. These best practices are likely to assume an even greater significance in Impact evaluation studies show that the program the wake of the COVID-19 pandemic, as countries is resulting in improved education and health out- attempt to mitigate the negative effects of the pan- comes among beneficiaries, including enhanced demic on human capital outcomes. 43 Eligible households received between 500 pesos and 1,400 pesos (US$11–US$32) per month, depending on the number of eligible children in the household (King 2020). 44 In 2009, the DSWD institutionalized the system as the National Household Targeting System for Poverty Reduction (NHTS-PR), and by 2011 it had shared the database with the Philippine Health Insurance Corporation, Department of Agriculture, and Depart- ment of Health to help those agencies better target the benefits of their own programs (Fernandez and Olfindo, 2011). 45 Acosta and Velarde (2015). 46 Gelli et al. (2019). 47 World Bank (2012) and Sumberg and Sabates-Wheeler (2011). 48 The GSFP is run by the Ghana School Feeding Program Secretariat under the direct supervision of the Ministry of Local Govern- ment and Rural Development. Other public partners directly involved include the Ministry of Education, the Ministry of Food and Agriculture, the Ministry of Health, the Ministry of Women and Children’s Affairs, the Ministry of Finance and Economic Planning, and the District Assemblies. 3 Accumulation Interrupted? COVID-19 and HUMAN CAPITAL T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 63 C OVID-19 has exacted a heavy toll in illness At the same time, lockdowns translated into school and lost lives and on the economy. Lacking closures and the shift to remote learning in some a vaccine or effective pharmaceutical treat- form, which can in many cases worsen learning gaps ment against SARS-CoV-2, the novel coronavi- between children with a more affluent background rus responsible for COVID-19, many countries and those who are less well off. It can also lead to wid- resorted to large-scale nonpharmaceutical inter- ening gaps between countries, since many may not ventions (NPI) to slow the virus’s spread. These have the infrastructure in place for such an endeavor. NPI resulted in an economywide lockdown of dif- Adding to people’s hardships are household income ferent levels of restrictiveness. These measures losses due to unemployment and reduced remit- further amplified the disruptions that COVID-19 tances, with effects that might be quite different brought to supply chains and global trade, add- across developed and developing countries.3 ing to the already dramatic economic dimension of the health crisis. A baseline forecast for GDP While there is still tremendous uncertainty on the in 2020 predicts a global drop of 5.2 percent,1 the overall impact of the pandemic on human capital, worst recession in eight decades, which is likely to it is clear that both direct and indirect pathways will push 100 million more people into poverty.2 matter. Those who were most vulnerable to begin with are likely to be the worst hit, and many dimen- A lesson from past pandemics and crises is that sions of inequality are likely to increase. Sections their effects not only are felt by those directly 1 and 2 of this chapter discuss channels of impact impacted, but often ripple across populations and from COVID-19 to human capital and their likely in many cases across generations. COVID-19 is no effects over people’s full life cycle. Section 3 dis- exception. Both the health and economic effects cusses how the Human Capital Index (HCI) can be of the disease and its control measures have sig- used to quantify some of the likely impacts of the nificant consequences for people’s human capital. pandemic on children and youth. In many health systems, the fight against the pan- demic has crowded out other essential health ser- vices. At the same time, people’s fear of infection 3.1 TRANSMISSION OF THE COVID-19 has led to many choosing to not seek treatment, SHOCK TO HUMAN CAPITAL possibly derailing years of gains against diseases like tuberculosis, HIV, and malaria. 3.1.1 Health system disruptions As governments scramble to respond to the imme- diate consequence of the pandemic, resources are 1 World Bank (2020). 2 Mahler et al. (2020b). 3 Simulations suggest that, in Ireland, 400,000 households may see a drop in their disposable income of 20 percent or more (Beirne et al. 2020). In Italy, simulations show that disposable income losses will be considerable and more pronounced for the poorest. Italian households in the poorest quintile are projected to lose 40 percent of their income (Figari and Fiorio 2020). 64 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l likely to be diverted from other health efforts that COVID-19 infection. Such service interruption will nonetheless remain critical. In past health emer- also likely lead to numerous deaths, many of them gencies, substantial negative indirect effects have avoidable. For example, in high-burden coun- resulted from this crowding out of nonpandemic tries, it is estimated that over the coming five years -related health services. For example, in the deaths due to tuberculosis, HIV, and malaria will 2014–15 Ebola outbreak in West Africa, closure of increase by 20, 10, and 36 percent, respectively.8 A health facilities, health worker deaths, and excess lesson is that, when determining how to reallocate demand placed on the health system led to further resources for pandemic response, special attention loss of lives. In Ebola-affected areas, it was reported must be given to maintaining coverage of key non- that maternal and delivery care dropped by more COVID interventions.9 than 80 percent, malaria admissions for children under the age of 5 fell by 40 percent, and vaccina- 3.1.2 School closures tion coverage was also considerably reduced.4 By the end of April 2020, schools were closed or partly closed in roughly 180 countries, although Some of these consequences are already apparent schools are now slowly reopening in many juris- for COVID-19. Vaccination programs in roughly 68 dictions.10 While the impact of school closures will economies have been interrupted due to the pan- depend on the effectiveness of mitigation from demic, and some 80 million children under the age remote instruction, closures will likely result in a of 1 year will go unvaccinated in low- and middle-in- slowdown and loss of learning, and an increased come countries as a result. Supply-chain break- 5 likelihood of school dropout, particularly for the downs combine with forced mobility restrictions most disadvantaged and for girls.11 under NPI to complicate overall access to vaccines.6 These human capital losses are not necessarily uni- Children and pregnant mothers are not the only formly distributed across the population. As chil- ones who will suffer from weakened service deliv- dren learn from home, social inequalities become ery capacities and curtailed access to services. more salient. The closure of schools could widen During a pandemic, most people are more reluctant already-existing gaps in education between chil- to seek medical care. During the SARS epidemic dren from better-off homes and those who come in Taiwan, China, people’s fear of infection likely from less well-off backgrounds, as poor house- led to sharp drops in demand for access to medi- holds’ access to technology and infrastructure is cal care across the board.7 Many patients suffering likely to be more limited. Additionally, learning from other illnesses will be unable to go for routine from home requires more inputs from parents, checkups, due to restricted movement and to avoid and some parents’ limited capacity to guide and 4 Elston et al. (2017). 5 WHO (2020a) and Nelson (2020). 6 Ibid. 7 See Chang et al. (2004). 8 See Hogan et al. (2020). These authors find that for HIV the largest impact is from interruption of antiretroviral therapy, for TB the impact is due to reduction of timely diagnosis and treatment, and for malaria it reflects the interruption of prevention programs. 9 Roberton et al. (2020) suggest that maintaining key childbirth interventions like parenteral administration of uterotonics, anti- biotics, anticonvulsants, and clean birth environments could lead to 60 percent fewer maternal deaths. Maintaining coverage of antibiotics for neonatal sepsis and pneumonia and oral rehydration solution for diarrhea would reduce child deaths by 41 percent. These results are likely contingent on modeling assumptions. 10 UNESCO (2020). 11 See Azevedo et al. (2020). Girls’ educational outcomes during a crisis tend to fall more so than those of boys. This is particularly the case if parents’ perception of returns on investments for boys are greater than for girls (Rose 2000) T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 65 support their children’s learning could exacerbate many people to cut back on expenses, which in inequalities. turn may trigger more businesses to close and more people to lose their jobs.17 The ensuing eco- Along with education, many children receive other nomic decline is likely to undo years of gains in the services through their schools. These include meal fight to eradicate extreme poverty. Accordingly, programs, which tend to benefit poorer children. the World Bank has projected an increase in inter- The suspension of school feeding programs could national extreme poverty for the first time since worsen food insecurity and malnutrition. The 1998.18 burden of making up the nutritional shortfall now falls on parents, many of whom are struggling eco- Closures and decreased economic activity result in nomically due to the pandemic.12 higher unemployment and income losses for many households. Households in countries that rely 3.1.3 Income effects, price effects, and food on remittances or seasonal migrants for income security report that contributions from these sources have The emerging literature on containment strate- fallen considerably, and many households report gies highlights the large benefits—in terms of lives that they expect to lose their remittances alto- saved and GDP losses averted—of testing and con- gether (see Box 3.1). The fall in household incomes tact tracing. While countries such as South Korea 13 is likely to affect the poor disproportionately, as and Iceland successfully implemented these strat- they often experience more fragile labor arrange- egies early on in the pandemic, most countries ments and, if inadequately covered by safety nets, resorted to lockdowns and movement restriction.14 are likely to fall through the cracks. Voluntary mobility restrictions combined with government-driven lockdowns generate a signifi- The income shock will probably be exacerbated by cant drop in activity and aggregate demand that the initial price shock already observed in many is leading to a considerable reduction in incomes. countries. The pandemic has created a short-run Nevertheless, the largest impacts to the economy demand shock, where the products demanded by are expected to come from reduced consumption consumers are different. As movement restric- due to people’s avoidance of social interaction due tions dissuade people from venturing out in pub- to fear of infection.15 lic, many activities that would typically happen in markets, restaurants, or other commercial settings Projections show that the resulting economic fall- end up taking place at home. Because manufac- out will be massive and potentially worse than that turing of goods for restaurants, hotels, and offices of the 2008–09 financial crisis.16 Lockdowns force differs from manufacturing for home consump- many nonessential businesses to close and will fur- tion, which has now increased, shortages can tem- ther disrupt supply chains. Coupled with inherent porarily arise and prices increase as a short-run uncertainty due to the pandemic, this may prompt response. 19 12 Lancker and Parolin (2020). 13 Acemoglu et al. (2020). 14 Hale et al. (2020). 15 Wren-Lewis (2020). 16 Ibid. 17 International Monetary Fund (2020). 18 Mahler et al. (2020a). 19 Hobbs (2020). 66 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l Concerns about localized food availability may not food expenditures of roughly 50 percent per adult be unfounded. Due to mobility restrictions, many equivalent. Evidence from Uganda also points farmers may experience labor shortages, which can toward temporary coping mechanisms used by reduce yields and further strain the supply of sta- households, many of which increased borrow- ple foods.20 Small farmers may also choose to avoid ing, dipped into their savings, or invested more of going to markets to sell their goods, due to fears of their time in household enterprises.25 contagion. Mobility restrictions and labor shortages may also prevent farmers from transporting their Despite the pandemic’s severe direct health goods to market. This is likely to affect the avail- impacts, the largest effects on human capital ability of more perishable crops, such as fruits and will probably come through indirect channels. vegetables. If these products cannot reach markets Indirect does not mean insignificant. Emerging in time, they may simply rot in the fields, as many results for a large set of rapid phone surveys farmers lack adequate storage facilities.21 fielded by the World Bank speak to indirect con- sequences of the pandemic that may perma- Given that many households will experience a nently weaken countries’ human capital for gen- fall in their incomes, many households will likely erations (see Box 3.1). experience food insecurity. This will impact the poorest households most, since they devote a larger share of their incomes to food expendi- 3.2 THE COVID-19 HUMAN CAPITAL tures. Households will respond to such events by SHOCK: A LIFE-CYCLE PERSPECTIVE limiting their food intake and/or relying more on cheaper staple foods, reducing dietary diversity. The accumulation of human capital is the result of This will further worsen the nutrition of millions a dynamic process whose dimensions complement of people. 22 Evidence of such scenarios is already each other over time. Depending on an individual’s emerging. For example, in Senegal, 86 percent stage in life, the impact of the pandemic on this of respondents to a telephone survey reported a process may come through different channels and drop in their incomes, and more than one-third have a differential impact. Setbacks during certain indicated that they restrict their meals four to stages of the life course—chiefly early childhood— seven days a week.23 In Nigeria respondents state can have especially damaging and long-lasting fear for their health and financial future, with effects. For example, economic hardship can force many also reporting increased prices of major families to prioritize immediate consumption food items and loss of employment. 24 In Uganda, needs, forgoing spending on health or education. households on average report a reduction in total Because demand for investing in human capital household incomes of 60 percent, and a drop in rises with incomes,26 a fall in incomes could worsen 20 This was observed during the 2014–15 Ebola outbreak in West Africa. See de la Fuente, Jacoby, and Lawin (2019). A similar effect is now seen in India, where nonavailability of migrant labor has interrupted harvesting activities. See Saha and Bhattacharya (2020). 21 Tesfaye, Habte, and Minten (2020). 22 Women will often sacrifice their own consumption needs in order to ensure sufficient nutrition for other members. See Quisumb- ing et al. (2011). 23 See Le Nestour and Moscoviz (2020). 24 Lain et al. (2020). 25 Mahmud and Riley (forthcoming). When surveyed, Ugandan households had yet to resort to selling productive assets to cope with the losses in income, perhaps in the hope that the income shortfall will be short-lived. 26 Bardhan and Udry (1999). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 67 Box 3.1: Rapid response phone surveys reveal immediate impacts of COVID-19 on the poor Although the impacts of the pandemic are cross-cutting, they are particularly damaging for the poor and vulnerable. Policy makers need timely and relevant information on the impacts of the crisis as well as the effectiveness of their policy measures to save lives, support live- lihoods and maintain human capital. To track the socioeconomic impacts of the pandemic, the World Bank rolled out the rapid response phone survey (RRPS) in more than 100 coun- tries. Traditional face-to-face surveys are hindered by social distancing protocols and mobility restrictions, while phone surveys overcome these limitations. They can be deployed rapidly, implemented at low cost, used to regularly collect longitudinal information, and adapted swiftly to changing circumstances. Preliminary results are available for Ethiopia, Kenya, Mongolia, Myanmar, Nigeria, Tajikistan, and Uzbekistan. With severe mobility restrictions imposed to limit the spread of the pandemic, economic activities have become severely disrupted. Many workers—especially in the service sector but also in agriculture, for example, in Myanmar and Ethiopi—lost employment. In Kenya, the unemployment rate tripled, while in Myanmar, Nigeria, Tajikistan, and Uzbekistan more than 1 in 5 households lost all employment. Despite some signs of recovery in employment, especially in Ethiopia, Nigeria, Tajikistan, and Uzbekistan, more than half of households report income losses in the remaining countries. Mainly due to the loss of income, food insecurity increased often substantially. It tripled in Nigeria and dou- bled in Tajikistan compared to the previous year. On the other hand, access to education has been severely limited in most countries, particularly for rural and poor households. In all coun- tries where data were analyzed so far, schools were closed and replaced with remote learning activities. While survey questions across countries are not strictly comparable, access and utilization of remote learning activities vary widely. Almost all children are engaged in remote learning activities in Uzbekistan, 7 out of 10 children are learning remotely in Mongolia and Kenya, 6 out of 10 in Nigeria, and only 3 out of 10 in Ethiopia and Tajikistan. The type of learn- ing activities also differ, for example, with Kenyan children mainly studying independently, while in Uzbekistan almost 9 out of 10 children report using educational television programs. In most countries, children living in rural or poor households are more affected by school clo- sures due to more limited access to remote learning. Access to medical services seems less affected, with 10 percent, over 16 percent, and 25 percent of households unable to obtain medical treatment in Mongolia, Uzbekistan, and Kenya, respectively. Source: World Bank Global Poverty team and https://www.worldbank.org/en/topic/poverty/brief/high-frequency- monitoring-surveys. human capital accumulation for many people, over the life cycle. On the right-hand side of the especially the most disadvantaged. 27 Figure 3.1 picture are some typical age-specific markers for depicts schematically how some of these shocks can human capital development, some of which enter affect the process of human capital accumulation as components into the Human Capital Index. 27 In some cases, the substitution effect (the relative change in prices of activities) dominates the income effect (the drop in purchas- ing power). For example, Miller and Urdinola (2010) present evidence of how child health has improved among children of coffee farmers in Colombia during a decline in the price of coffee. Since time spent farming is less valuable due to the fall in coffee prices, parents devote more time to their children, which translates into better outcomes for children. Schady (2004) provides evidence that, in Peru, children exposed to a crisis in the late 1980s completed on average one additional year of schooling. 68 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l Figure 3.1: Human capital accumulation across the life cycle—key stages and metrics Stunting School attainment and learning; School Unemployment; Years of healthy life Child mortality; Worse health Markers attainment and expectancy; Increased Low birth weight learning morbidity and mortality College In utero Birth 0–5 5–18 18–60 60+ Working Life Shock Dropout and learning Displaced Morbidity, care at birth decline due to school stress, and closure and lost isolation income Mother’s Possible Unemployment malnutrition due malnutrition and drop in income to lockdown 3.2.1 From conception to age 5 lower educational attainment and income during During childhood, the link between parental adulthood. The effect was even more salient income and child health is particularly strong. 28 among children of infected mothers.32 Much For example, reduced nutrition in pregnant moth- about the current virus remains to be learned. ers could have a substantial impact on children in At the moment, the main transmission channel utero, including long-lasting impacts on chronic affecting the fetus’s human capital is expected to health conditions and cognitive attainment in be through the disruption of health care and lower adulthood.29 The evidence shows that this is the household income.33 case for children born during a pandemic but also for children born during conflict30 and eco- Birthweight is often interpreted as a key observ- nomic hardship. For example, children who were 31 able component of a child’s initial endowment.34 in utero during the 1918 influenza pandemic had Children who were in utero during the 2008 28 See Almond (2006). 29 See Almond and Currie (2011). 30 For example, Bundervoet and Fransen (2018) find that children exposed to the Rwandan genocide while in utero suffered lower educational outcomes. The longer the exposure in utero, the poorer the educational outcomes. 31 Rosales-Rueda (2018). 32 See Almond (2006). 33 Savasi et al. (2020) found that 12 percent of the 77 patients in their study (in Italy) had a preterm delivery. On the other hand, Philip et al. (2020) find a reduction in preterm births in Ireland, and a reduction in very low birth weights, falling from 3.77 cases per 1,000 births to 2.17 cases. 34 See Datar, Kilburn, and Loughran (2010). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 69 recession were born with relatively lower birth- child mortality, with a more marked increase in weight, particularly in families at the bottom of lower-income countries. A meta-analysis of studies the income distribution.35 This was the case for for developing countries suggests that a 10 percent children born in those California regions that suf- increase in GDP per capita is related to a decrease fered unusually elevated unemployment rates in infant mortality of 4.5 percent.40 Recent esti- after the 2008 recession.36 Similarly, in Ecuador mates also show that the relationship between during the 1998 El Niño floods, children who were income and child mortality is likely higher in in utero and especially in the third gestational low-income countries, suggesting that short-term trimester were much more likely to be born with aggregate income shocks translate into an increase low birthweight, and these children showed sub- in child mortality of 1.3 deaths per 1,000 children stantially reduced stature 5 and 7 years afterward.37 among low-income countries, given a 10 percent These health effects were attributed to drops in decrease in per capita GDP.41 household income following the devastation of El Niño. Similar outcomes can unfortunately be Stunting rates are also likely to increase due to the expected from the COVID-19 shock. As low birth- COVID-19 shock. Common factors related to stunt- weight is associated with increased likelihood of ing are maternal nutrition during pregnancy and malnutrition and developmental delay, COVID- nutrition during infancy, both of which will likely 19-induced income effects may substantially worsen if families have less disposable income.42 affect human capital attainment for generations A fall in aggregate GDP could also lead to weak- to come. 38 ened health infrastructure and less funding for nutritional interventions and services.43 Existing Child mortality is unfortunately also likely to estimates of elasticities suggest that a 10 percent increase, for two reasons. The first is the disrup- increase in GDP leads to a decrease in stunting that tion in maternal and child health services due may range from 2.7 to 7.3 percent.44 Nevertheless, to COVID-19. Early simulated values project an aggregate elasticities may obscure the fact that increase of child mortality of up to 45 percent many of these shocks will affect the poor and dis- due to health-service shortfalls and reductions advantaged disproportionately. Attention must in access to food in 118 low-income and middle- be paid to ensure these groups have access to any income countries.39 Second, economic downturns available support mechanisms that may mitigate have been associated with significant increases in such impacts. 35 See Finch, Thomas, and Beck (2019). 36 Ibid. 37 See Rosales-Rueda (2018). 38 See Black, Devereux, and Salvanes (2007) and Lahti-Pulkkinen et al. (2018). 39 Roberton et al. (2020). 40 O’Hare et al. (2013) obtain this estimate through meta-analysis from a systematic literature search of studies and find a pooled elasticity of income on infant mortality of −0.95. 41 Ma et al. (2020) find that, in low-income countries, a lockdown will potentially lead to 1.17 children’s lives lost per COVID-19 fatality averted, due to the economic contraction, significantly higher than in lower- and upper-middle, income countries (where it would stand at 0.48 and 0.06, respectively). This is due to two factors: the younger demographic structure and the higher estimated elas- ticity of child mortality to GDP changes in low-income countries. The authors also assume that under-5 mortality is not affected by income shocks in high-income countries. 42 See Galasso and Wagstaff (2019). 43 See Mary (2018) for a more nuanced discussion. 44 Mary (2018) suggests that the decrease may be 2.7 percent, while Mary et al. (2019) estimate it to be 7.3 percent, and Ruel et al. (2013) suggest 6 percent. It is worth noting that these analysis concentrated mostly on low- and middle-income economies. 70 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l 3.2.2 The school years Tanzania, among agricultural households, income With almost all countries having imposed some shocks, even transitory ones, led to increased child type of school closure due to the pandemic, stu- labor and reduced school attendance.47 dents in many settings are likely to suffer learning shocks. Evidence suggests that any interruption Evidence from natural disasters confirms that in children’s schooling typically worsens learn- interruptions and trauma in the neurodevelop- ing outcomes. This includes disruptions caused mental process can adversely affect academic per- by epidemics, conflict, natural disasters, and even formance.48 Four years after bushfires in Australia, scheduled school vacations. US students’ achieve- children from areas that were heavily impacted by ment scores appear to decline by about a month’s the fires performed worse in reading and numer- worth, on average, during the regular three-month acy than peers from less-impacted schools.49 The summer break.45 case of the bushfires underscores the importance of continued support to affected populations, since Historical experiences illustrate the impacts of a longer-term learning divergence was found even large-scale school closures during a public health though students did not display any differences in emergency. Meyers and Thomasson (2017) stud- learning outcomes immediately after the disaster. ied the effects of the 1916 polio pandemic on edu- cational attainment in the United States. Young Further indication of the damage caused by people ages 14–17 during the pandemic later school interruptions can be gleaned from the showed reduced overall educational attainment outcomes after the 2005 earthquake in Pakistan. compared to slightly older peers.46 Even short- Areas near the fault line were devastated, 80 term school closures appeared to have lasting percent of homes were destroyed, and schools impacts on children’s educational attainment, suffered considerable damage. Cash transfers though the study found such effects only among played an important mitigating role, because children who were of legal working age during the four years after the earthquake, households school closures. near the fault line were indiscernible, in welfare terms, from those farther away from the fault Increased dropout rates are one relay linking line. Enrollment rates for children residing near emergency school closures to future losses in life- the fault line were not affected. However, despite time educational attainment. In general, as chil- the apparent return to “normalcy,” test scores for dren age, the opportunity cost of staying in school children living 10 kilometers away from the fault increases. This may make it harder for households line were 0.24 SD below those of children residing to justify sending older children back to school 40 kilometers away.50 after a forced interruption, especially if house- holds are under financial stress. Again, such effects Many countries have adopted distance learn- are not restricted to public health emergencies. In ing as a means to mitigate learning losses during 45 Cooper et al. (1996). More recent research has called this result into question. See von Hippel and Hamrock (2019) for more nu- anced discussion. However, a summer break is not the same as a break during the school year. 46 Meyers and Thomasson (2017). 47 Beegle, Dehejia, and Gatti (2006). 48 Gibbs et al. (2019). 49 Ibid. 50 Andrabi, Daniels, and Das (2020). The authors posit that this is equivalent to 1.5 school grades. To arrive at this value, the authors note that the average 15-year-old has accumulated 5.6 grades and linearly gain 0.17 standard deviations (SD) in performance per grade level in the test the authors use. This result, in the context of harmonized test scores used in the HCI, translates to a drop of 24 points. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 71 protracted school closures. Remote teaching strat- system. This occurred despite the government’s egies include not only online learning, but also offering of a 25 percent subsidy on monthly pri- radio and TV programs and text nudges in those vate-school tuition for parents who had lost jobs.54 countries where digital connectivity is limited. With limited numbers of qualified teachers avail- These strategies make it less likely that negative able, migration of students from private to public effects of similar magnitude to other interruptions schools could worsen learning outcomes across will be replicated; however, the effectiveness of many countries. these measures has yet to be determined. Thus, the impacts of school closures extend far The most recent global projections on the impact beyond initial enrollment drops. For girls, school of school closures linked to COVID-19 suggest closures may also lead to increased exposure to that, using the HCI metric of learning-adjusted pregnancy and sexual abuse. In many countries years of schooling (LAYS; see Box 1.1 from Chapter this could be worsened by policies that prevent 1), almost 0.6 years will be lost due to the closures. "visibly pregnant girls" from attending school.55 These numbers reflect the loss of schooling that Both shorter- and longer-term impacts are likely comes from potential dropouts due to the loss of to affect disadvantaged families most, further wid- income,51 as well as the adjustment in quality due ening inequalities in learning and human capital to worsened learning because of inefficient remote accumulation between socioeconomic groups. teaching methods.52 The lost schooling in the face of a mitigation strategy that has medium efficiency Finally, a drastic change in the day-to-day lives translates to a yearly loss of over US$ 872 in 2011 of children and adolescents is likely to affect USD PPP, reaching a loss of US$ 16,000 in lifetime their mental health. The pandemic may worsen earnings in present value terms at a discount rate already-existing mental health issues by provok- of 3 percent and assuming a 45-year work life. 53 ing or exacerbating social isolation, economic As children head back to school, countries with uncertainty, and fear.56 A recent study among an already overextended education system may Ecuadorian teenagers (ages 14 to 18) found that one be grappling with increased demand for public in six teenagers reported suffering from depres- education. This has occurred due to household sion, while many cited household finances and income losses that have prompted many parents social isolation as concerns.57 The use of digital to turn to public schools. In June 2020, registration technology, particularly with voice and video, can in public schools in the coastal zone of Ecuador, ameliorate the loneliness faced by many teens and for example, increased by 6.5 percent, bringing children, but these technologies are not available some 120,000 additional students into the public to all.58 51 The simulation by Azevedo et al. (2020) implicitly assumes that income effects outweigh substitution effects that may arise in these cases. Nonetheless substitution effects may be larger, and enrollment could increase. Shafiq (2010) presents two cases: (1) Falling wages make child labor less attractive, and (2) if parents place a higher preference on education, perhaps because less educated workers bear the brunt of the crisis, then enrollment may increase. 52 Azevedo et al. (2020). 53 Ibid. Values are obtained for 157 countries. Authors model different mitigation strategies taken during remote learning, vary the length of school closures, and assume children will drop out of school due to the income shock. The yearly losses range between US$ 127 in low-income countries to US$ 1,865 in high-income countries per year. 54 Olsen and Prado (2020). 55 Bandiera et al. (2019). The determination of correctly identifying pregnancy gave school principals discretion on how to enforce the ban. 56 Golberstein, Wen, and Miller (2020). 57 Asanov et al. (2020). 58 Galea, Merchant, and Lurie (2020). 72 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l 3.2.3 School-to-work transition and tertiary the average effect hides substantial heterogeneity. education Recent graduates with the lowest predicted earnings The pandemic is also disrupting human capital are likely to suffer the largest losses and often do not accumulation for students currently in tertiary recover the lost ground after 10 years.62 Starting at a education. Almost the totality of students currently lower-paying job or at a less-desirable firm that does enrolled in tertiary education are experiencing a not make full use of an individual’s existing human new learning modality.59 With students in low- and capital may well lead to a lag in skill accumulation middle-income countries less likely to have inter- and result in a persistent disadvantage. net access, between-country inequalities in learning will worsen. Within countries, those at the bottom of Women who graduate from high school during the income distribution will also be more affected, the pandemic may choose to respond differently due to lack of access to the necessary materials for than male peers to the shock and forgo college in remote learning. This will again exacerbate existing the short term. They are also less likely to join the inequalities in human capital accumulation. workforce, due to the depressed wages. Evidence from the United States suggests that women, but not Two opposing forces may influence tertiary enroll- men, graduating from high school are more likely to ment rates. Pandemic-induced high unemploy- skip college during recessions because of the lower ment rates are likely to reduce the opportunity cost observed returns to education and because the cost of attending college. At the same time, the recession of more schooling increases.63 For some, the alter- will affect many households economically, and funds native of child rearing may be more attractive in for attending college may not be available. After the the short term, as was the case during the 2008–09 financial crisis, enrollment rates for tertiary educa- global recession. For others, disruptions in the supply tion in the United States went up. However, because chain may lead to unintended pregnancies as many of a substantial decrease in family incomes, student women will lose access to modern contraceptives.64 shifted away from four-year private colleges toward two-year public institutions.60 Finally, because of the depressed wages and fewer legal employment options available during a Those who graduate from college now are also likely to recession, crime also becomes more attractive. suffer short- to medium-term wage losses. Evidence The longer the recession lasts, the more likely from Canada suggests that graduating during a reces- that acquired human capital depreciates and sion is linked to significant initial earning loss due crime becomes a worthwhile option. This effect is to less desirable job placements, but that this pen- heightened for those who have lower human capi- alty fades over some 8 to 10 years.61 Nevertheless, tal levels and are less attached to the labor market.65 59 Bassett and Arnhold (2020). 60 Dunbar et al. (2011). A similar dynamic was observed in Peru, where the opportunity costs of going to school decreased by a con- siderable amount because wages dropped substantially. Thus, children exposed to the crisis completed more years of education (Schady 2004). 61 Oreopoulos, von Wachter, and Heisz (2012). 62 Rothstein (2020) finds evidence that those who graduated during the 2008–09 financial crisis had lower wages and employment than earlier cohorts. The author shows that market conditions at the time of labor market entry matter greatly for cohorts’ employ- ment probabilities. 63 Hershbein (2012). 64 Roughly 47 million women in 114 low- and middle-income countries could lose access to contraceptives in the scenario of a six- month lockdown or disruptions (United Nations Population Fund [UNFPA] 2020). 65 Bell, Bindler, and Machin (2017) find that cohorts that graduate into a recession are 10.2 percent more likely to commit criminal activity than cohorts who enter the labor market in nonrecession times. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 73 3.2.4 Working life There is also evidence that those who have lost Together with the economy, the pandemic has jobs during the pandemic could suffer far more affected labor markets dramatically. According than lost earnings. One study finds that US work- to the International Labour Organization (ILO), ers who were employed in a firm for at least six working hours during the first quarter of 2020 years and were then dismissed during a recession declined by the equivalent of 130 million full-time had higher mortality rates than similar workers jobs. The organization expects that the results will who had not been displaced. The estimates sug- be even worse in the second quarter of 2020, with gest an average decrease in life expectancy for the the number climbing to 305 million full-time dismissed workers between 1 and 1.5 years, likely jobs. 66 The pandemic and lockdown measures are due to increased chronic stress. Even when eco- affecting workers worldwide but are having par- nomic conditions later improved, the lower earn- ticularly dramatic impacts for informal workers. ings experienced by workers who lost jobs can lead Informal work often happens in crowded places, to reduced investments in health.71 so that lockdown measures—when enforced strictly—make continuing with these jobs impos- The pandemic and the nonpharmaceutical inter- sible. 67 Informal workers also often fall through ventions taken are also likely affecting women the cracks of social protection systems, lacking more than men. The sectors typically most access to unemployment and health insurance. 68 affected by lockdowns have high shares of female employment.72 School closures will likely contrib- Unemployment stints, even short ones, tend to ute to heavier workloads for many women, mostly leave a lasting mark on individuals’ earnings. because women are likely to be responsible for child For many, this will be the second “unprece- care in the absence of alternatives. These pressures dented” economic shock of their working lifetime. may limit women’s paid work.73 Established gen- Workers who have longer tenures in a company, der norms are also likely to prevail when a family if dismissed, are likely to face a considerable ero- member falls ill due to COVID-19, with women in sion of skills, as many skills they have accumu- the household expected to care for the sick. On the lated may be particular to that employer. If these other hand, the current shift to flexible working workers find employment in the future, and their arrangements could benefit some workers, includ- new job requires different skills, they are likely to ing women, and could promote gender equality in experience a considerable wage penalty. 69 Those the labor market in some settings.74 who lose a job during a mass layoff event are likely to experience large and persistent earning losses, Beyond work, interpersonal violence is also on roughly equivalent to 1.7 years of their earnings the rise, leaving many women more exposed due prior to dismissal. 70 to the lockdown.75 Evidence of this has already 66 International Labour Organization (2020a). 67 Ibid. 68 Packard et al. (2019). 69 Poletaev and Robinson (2008). 70 Davis and von Wachter (2011). 71 Sullivan and von Wachter (2009). 72 Alon et al. (2020). 73 Wenham, Smith, and Morgan (2020). 74 Alon et al. (2020). 75 Gelder et al. (2020). 74 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l surfaced. For example, in Argentina the lockdown vulnerabilities revealed by COVID-19 point to the restrictions were directly linked to an increase need to increase human capital resilience. This in calls to the domestic violence hotline of 28 could mean rethinking policies and services for percent. Additionally, also in Argentina, women today’s elders, but also supporting younger gen- whose partners were also in quarantine were more erations to prepare for a healthy longevity in the likely to report an increase in interpersonal vio- future. This will involve stepping up prevention of lence due to increased exposure to the perpetra- noncommunicable diseases such as cardiovascu- tor. There also is evidence of this in India, where 76 lar diseases, obesity, and diabetes, in conjunction domestic violence complaints increased most in with other strategies. regions that implemented a more strict lockdown. 77 3.2.5 Older adults 3.3 USING THE HCI TO SIMULATE THE The risk of adverse health effects from COVID- IMPACT OF THE PANDEMIC 19 increases significantly with age and comor- bidities, making the elderly especially vulner- The HCI is designed to capture the human capital able. Residing in a long-term care facility also a child born today can expect to attain by age 18. substantially increases risk. For example, pre- Given that the future is uncertain, the best approx- liminary analysis of April 2020 COVID-19 expo- imation of human capital accumulation for a child sure data in Italy indicated that 44 percent of born today is based on the currently observed out- infections during this period were contracted in comes of older cohorts. While there is uncertainty nursing homes or homes for the disabled.78 In about how long it will take for the world to arrive the United States, as of mid-May 2020, nurs- at a post-COVID-19 “new normal” (and what the ing-home residents accounted for about one- world will look like then), for the purpose of the third of COVID-19 fatalities. 79 While such find- long-term outcomes captured by the HCI, the ings are alarming, they probably underestimate pandemic is mostly a transitory shock. For exam- actual infection and case fatality rates among ple, while school closures affect school-age chil- older adults, since there is evidence that, espe- dren now, these are unlikely to affect children who cially at the beginning of national epidemics, are born today, assuming that the pandemic will deaths from COVID-19 went unrecorded in be controlled and school will be in session by the many long-term care facilities. time they are ready to start school. An immediate priority for countries fighting However, the disruption to health systems and COVID-19 is to protect the elderly and those shocks to family income will affect young chil- with significant comorbidities. Prevention, con- dren’s survival and healthy development (stunt- trol, appropriate staffing, coordination, manage- ing) now. In turn, this will affect their learning and ment, reporting, communication, and planning schooling. Since all the data for the 2020 HCI were are all needed to safeguard older adults living collected just before the virus struck, it serves as in residential facilities. 80 In the longer run, the a pre-COVID-19 baseline, and the HCI construct 76 Perez-Vincent et al. (2020). The authors of the study also note a considerable increase in the number of calls related to psycholog- ical violence, by 57 percent. 77 Ravindran and Shah (2020). 78 Task force COVID-19 del Dipartimento Malattie Infettive e Servizio di Informatica, Istituto Superiore di Sanità (2020). 79 See CDC (2020) and Yourish et al. (2020). 80 Such facilities include long-term care homes, residential care homes, nursing homes, welfare homes, and others. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 75 can be used to simulate the direct and indirect intertwined, the shock is also expected to affect the impacts of the pandemic on young children’s amount of education this cohort of children will human capital.81 Over time, it can be used to track attain in the future as well as how much education the actual changes in human capital outcomes as they can retain. The calculation of the impact is the pandemic evolves. developed formally in Annex 3B of this chapter. The rest of this section discusses an example of a Take a reduction in GDP per capita of 10 per- simulation of the effects of the pandemic shock on cent—a pessimistic scenario—with an elasticity the future human capital of young children under of stunting to income of –0.6, this would imply 5 years of age. Then it uses the HCI to simulate an increase in stunting of 6 percent. For example, how the pandemic—through school closures and for a country like Bangladesh, where the stunt- shock to family income—will affect the future ing rate pre-COVID-19 was 31 percent, an income human capital of children who are currently in shock of this magnitude could increase stunting school. by 1.85 percentage points.83 Since children who are stunted are less likely to stay in school and 3.3.1 Shock to children under 5 learn, this increase in stunting could lead to a While COVID-19 is seemingly not as damaging to drop in expected years of school of 0.03 years, the health of children or pregnant mothers who and a drop in harmonized test scores (HTSs) of are directly affected by it as previous pandemics,82 1.16 points.84 With about 10 years of schooling and the COVID-19 economic shock is expected to be an HTS of 370, the losses due to an increase in harmful for the youngest children and children stunting could amount to nearly 1 percent of the in utero, because considerable drops in family HCI. Adding in the likely increase in child mor- income can lead to food insecurity, in turn leading tality due to health service disruptions and the to increased child mortality and stunting. An addi- income drop would further drive down the HCI tional shock is the decrease in coverage of essen- by an additional 0.10 to 0.47 percentage points, tial health interventions for pregnant mothers and depending on the assumptions. Altogether, a young children. This decreased coverage is the decline in income of 10 percent could lead to a direct result of disruptions to health systems due decline in the HCI ranging from 1.13 to 1.50 per- to the health workforce and supply chain issues. cent. For a country like Bangladesh with an HCI These shocks will affect child mortality and child score of 0.46, this would imply a decline to an health. Mapping them into changes in human cap- HCI score of around 0.45. ital as measured by the HCI requires estimates of how substantially mortality and stunting change in Annex 3A to this chapter reports the methodology response to shocks to GDP per capita, as well as of the simulation in more detail. For each country, reductions in health services. the percentage decline in income due to COVID-19 is estimated as the difference between projections Since child health (captured by worsened stunt- of per capita GDP growth made in June 2020 and ing rates) and educational outcomes are closely the pre-COVID-19 projections made in late 2019. 81 See appendix A for details on the methodology of the HCI. 82 Almond and Currie (2011). In early 1919, roughly one-third of all newborns had mothers who had been infected by influenza while pregnant. The 1918 pandemic was disproportionately deadly to those between 25 and 35 (Almond 2006). 83 Assuming an elasticity of −0.6, where a 10 percent drop in GDP per capita would increase stunting by 6 percent. 84 These calculations are based on the literature review in Galasso and Wagstaff (2019). The authors find that children who are stunted obtain 1.594 years less education, and score 0.625 standard deviations lower on standardized tests. 76 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l Table 3.1: Simulated drop in HCI due to the pandemic’s impacts on children 5 and under Percentage point difference between June HCI 2020 % drop in HCI 2020 and AM19 projections By World Bank income group High income 0.707 −0.17 −6.23 Upper-middle income 0.560 −0.42 −7.77 Lower-middle income 0.480 −0.64 −5.36 Low income 0.375 −0.73 −4.34 Global 0.561 −0.44 −6.16 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: Calculations are based on the methodology presented in Annex 3A. Projected GDP changes are from the June version of the Global Economic Prospects and the 2019 Macro Poverty Outlook (Annual Meetings of 2019 [AM19]). The simulation assumes a four-month interruption to health system access. The access scenario comes from Roberton et al. (2020) and assumes considerable reductions in the availability of health workers and supply due to pandemic-driven reallocations. The scenario also assumes reduced demand due to fears of infection and movement restrictions as well as economic pressure. The scenario from Roberton et al. (2020) does not include additional child deaths due to wasting. It then applies the calculations described above 3.3.2 Schooling and learning87 to simulate the likely effects on human capital as The human capital of the current cohort of school- measured by the HCI for each country. Averaged age children is being heavily impacted by the pan- across all countries, the projected shock would demic through school closures: at its peak, nearly result in an HCI loss of 0.44 percent. This out- 1.6 billion children worldwide were out of school. come is worse for low- and lower-middle-income The simulation framework proposed by Azevedo countries (−0.73 and −0.64 percent, respectively), et al. (2020) quantifies the effects of this shock on mostly because the stunting rates are highest for the global stock of schooling and learning through this group of countries (Table 3.1).85 While the loss three channels: during closures children lose out may not seem large, it will likely set back children on opportunities to learn, they may forget what within the affected cohort for years to come, lead- they have previously learned, and many may drop ing to accumulated losses. For example, the cohort out due to income losses. of adults who were in utero during the 1918 influ- enza pandemic by 1960 had 0.1 less years of edu- According to the 2020 HCI, pre-COVID the global cation than those adults born in the year before or average of the expected years of school is 11.2 after the pandemic. When comparing the wages of years, which, when adjusted for learning, trans- the same groups in 1960, those who were in utero lated into 7.8 learning-adjusted years of school had wages that were 2.2 percent lower than those (LAYS). To simulate the effect of closures on learn- of the neighboring cohorts.86 These losses build up ing, the simulation starts by assuming a value of over time and leave affected cohorts at a consider- learning gained in one year of schooling. This is able disadvantage. proxied by harmonized test score (HTS) points per 85 For countries where stunting data are not available and thus not used in the calculation of the HCI, the income group’s average stunting rate is applied to the individual country to simulate the possible losses due to the pandemic. 86 Almond (2006). 87 This section was contributed by Joao Pedro de Azevedo based on Azevedo et al. (2020). The results presented in this section use the 2020 HCI numbers as baseline values. For that reason, they will be slightly different from those in the original paper. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 77 year.88 To determine how much of this will be lost months out of a 10-month school year, without any due to the closures, the simulation assumes three mitigation, students would lose 0.3 years of school. scenarios: optimistic, intermediate, and pessimis- Assuming instead, in the optimistic scenario, a mit- tic, corresponding to 3, 5, and 7 months of school igation effectiveness of 0.4, only 60 percent of that closures, respectively. The three scenarios also period would be lost, leading to a loss of only 0.18 differ on the assumed effectiveness of the mitiga- school years (0.3*(1-0.4)). A further dimension of loss tion measures put in place by governments, which comes from the drop in learning. Children in upper- vary by income group. 89 The three components middle-income countries like Peru gain 40 HTS are used to project HTS points lost due to school points in a school year. With students missing out closures under the assumption that the losses due on 0.18 years of school, they are also losing 7.2 HTS to school closures are not recuperated.90 points (40*0.3*(1-0.4)). Putting it all together means a loss of 0.27 learning-adjusted years of school. Expected years of school (EYS) are also projected to fall. This is because, due to the income shock, many Figure 3.2 depicts the combined losses in learning children are likely to drop out of school.91 COVID- and expected years of schooling for different coun- 19 could lead an additional 6.8 million children to try income groups. Under the intermediate sce- drop out of school around the world. Sixty percent nario of a five-month closure, COVID-19 could lead of these dropouts will be children between 12 and to a loss of 0.56 years of school, adjusted for quality. 17 years of age, who are likely to leave school per- This means that school closures due to COVID- manently because of losses in household income. 19 could bring the average learning that students The economic recession brought on by COVID- achieve during their lifetime down to 7.3 learn- 19, which is expected to shrink GDP per capita by ing-adjusted years of school.93 In the optimistic sce- 4 percent, is likely to increase the out-of-school nario, the projected loss is 0.25 years of schooling, population among global youth by 2 percent. and in the pessimistic scenario, 0.87 years. Take, for example, a country like Peru, which is an Across the globe, the extent of this shortfall will vary. upper-middle-income country, and assume a drop In high-income countries, where children were in GDP per capita of 10 percent. The drop in GDP expected to complete 10.3 years of learning-ad- is likely to lead many kids to drop out of school, justed schooling prior to the pandemic, the simu- leading to a small loss in expected years of school lations suggest that COVID-19 could lower LAYS to (0.005).92 An additional loss to years of school is due 10.1 in the optimistic scenario and 9.2 in the pessi- to the closures and the limited capacity of countries mistic scenario. At the other end of the spectrum, to deliver education during school closures. Under children in low-income countries were expected to the assumption that schools are closed for three complete 4.3 years of learning-adjusted school prior 88 For high-income countries, the value is assumed to be 50 points in a year; 40 points for upper-middle-income countries; 30 points for lower-middle-income countries; and 20 points for low-income countries. 89 The authors assume that all governments offer some alternative learning modality. Estimates of their effectiveness are informed by existing multitopic household surveys. Thus, access and effectiveness of the implemented modalities differ by country income. Efficiency for lower-, lower-middle-, upper-middle-, and high-income countries under the pessimistic scenario is 5, 7, 10, and 15 percent, respectively. The values are doubled in the intermediate scenario and quadrupled under the optimistic scenario. 90 The outcome results from multiplying the HTS points per year by the share of the school year that is assumed to be lost, and 1 minus the efficiency of the mitigation measure in place. 91 The authors use household surveys for 130 countries to calculate country-specific dropout income elasticities and welfare using cross-sectional variation by welfare quintiles. Refer to Azevedo et al. (2020) for details. 92 See Azevedo et al. (2020) for a detailed explanation on how the income shock is incorporated. 93 Intermediate scenario in Azevedo et al. (2020). 78 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l Figure 3.2: Learning-adjusted years of schooling lost due to COVID-19 school closures and income shock High income Upper-middle income Lower-middle income Low income Global -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0 Learning adjusted years of school lost Pessimistic Intermediate Optimistic Source: Azevedo et al. (2020). Notes: Results based on latest available LAYS for 174 countries (unweighted average). Coverage of 99 percent of the population ages 4–17 using the methodology from Azevedo et al. (2020). Projected GDP changes are from the June version of the Global Economic Prospects. School closure length: pessimistic 7 months, intermediate 5 months, and optimistic 3 months. Mitigation effectiveness also differs by scenario and income group. Refer to Azevedo et al. (2020) for full details. Table 3.2: HCI shock to children currently in school during the pandemic % drop in HCI if GDP for all countries % drop in HCI dropped by 10% By World Bank income group High income −5.17 −5.34 Upper-middle income −4.71 −5.04 Lower-middle income −4.00 −4.54 Low income −3.07 −3.66 Global −4.45 −4.82 Source: World Bank calculations based on the 2020 update of the Human Capital Index and on Azevedo et al. (2020). Notes: Calculation is based on the method presented in Annex 3.B. Projected GDP changes are from the June version of the Global Economic Prospects. to COVID-19. The optimistic scenario suggests that What is known about the virus itself continues this would fall to 4.1 years, while the more pessimis- to evolve, so many behavioral patterns are diffi- tic scenario foresees a decline to 3.8 years. cult to predict. For instance, parental concerns about child and family safety will likely dominate Putting these losses in LAYS in the context of the household decision-making around sending chil- HCI implies a drop in human capital for children dren back to schools when they reopen. Hence, of school age of 4.5 percent (Table 3.2). any estimates of dropouts that only consider T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 79 Table 3.3: Long-run lost human capital (HCI points) HCI points lost under GDP change from HCI points lost if GDP for all countries GEP (June 2020) dropped by 10% By WB income group High income −0.011 −0.012 Upper-middle income −0.009 −0.009 Lower-middle income −0.007 −0.008 Low income −0.005 −0.006 Global −0.0084 −0.0093 Source: World Bank calculations based on the 2020 update of the Human Capital Index and on Azevedo et al. (2020). Notes: Calculations are based on the methodology presented in the Annex 3.A and 3.B. Projected GDP changes are from the June version of the Global Economic Prospects (GEP). the relationship between incomes and school As an example, assume that a country’s HCI for dropouts are likely to underestimate the extent children under 5 is expected to fall by 1 percent to which children’s schooling and learning will and that they represent 15 percent of the work- be affected by the pandemic. Additionally, these force in 2040. Also assume that the losses due to numbers ignore the possibility of remediating school closures are 4 percent and these children these losses. will be 30 percent of the workforce by 2040. If the HCI in 2010 for this country was 0.54 and in 2020 3.3.3 The long-run HCI losses to the cohort it is 0.56, then the HCI of that country’s workforce In 20 years, roughly 46 percent of the workforce in 2040 will be 0.007 lower than it would have in a typical country (people ages 20 to 65) will be been in the absence of the pandemic.96 composed of individuals who were either in school or under the age of 5 during the COVID-19 pan- Given that children who are currently in school demic. 94 Assume that the 2020 HCI summarizes will be a larger share of the workforce than those well the human capital children under the age of 5 currently under 5, and that the losses for the for- could have achieved, and that the 2010 HCI is the mer are larger in high-income countries, the best representation of the human capital children fall is expected to be largest among high- and who are currently in school could have achieved. upper-middle-income countries, which are also With the HCI losses as calculated in the earlier sec- the ones that have the highest levels of HCI and tions, the HCI of the workforce in 20 years' time thus are projected to lose more (Table 3.3). The in the typical country would be lower by almost 1 results shown here are meant to inspire action and HCI point (0.01) due to COVID-19 today. 95 show that without remediation, an entire genera- tion could be left behind. 94 Roughly 34 percent of the workforce will be composed of individuals whose schooling was interrupted by the pandemic, and 12 percent of the workforce will be composed of individuals who were under the age of 5 during the pandemic. 95 To calculate the HCI loss by country, the percentage change for each of the cohorts (presented in Table 3.1 and Table 3.2) are applied to the HCI of 2020 for the under-5 cohort and the HCI of 2010 for the cohort of 2010 to arrive at an HCI value that is lost. Coun- tries missing an HCI in 2010 were imputed the value of the country’s income group. The country’s projected population shares are used to calculate each country’s HCI point loss among the workforce. The result supposes that on average those who are currently between ages 18 and 45 will not experience any ill health effects due to the pandemic, and thus in 20 years their human capital will be the same. HC ​I​loss​​ = 0.56​(− 0.01)​0.15 + 0.54​(− 0.04)​0.3​. 96 ​ 80 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l 3.4 ANNEX 3A: COVID-19 SHOCK TO THE UNDER-5 COHORTS The starting point for this simulation is a version of the HCI calculated with stunting only.97 Assuming that changes in stunting are sufficient to capture the health component of the index, through its relation- ship to height, and that the changes in adult survival rates due to the pandemic are unlikely to reflect the current health status of the current cohorts, the relevant HCI equation in log terms can be written as: ​lnHCI = ln​(Survival)​+ϕ​(EYS × ​_ 625 ​− 14)​+ ​γstunting HTS ​ ​(− stunting rate)​ The changes in HCI for the under-5 cohorts are assumed to come from the income shock and reduction in health care access during the pandemic. Consequently, the pathways for the shock are as follows: (1) The income shock affects under-5 mortality rates. (2) The income shock also leads to an increase in stunting rates. In turn, a change in stunting rates is expected to be related to a change in the years of school completed by affected children. It is also likely to be related to a change in cognition, proxied by harmonized test scores. (3) An additional shock is due to the reduced access to health services, be it due to fear of contagion or from the lockdown measures. This shock is expected to mostly affect under-5 mortality. Therefore, the fall in access to health services and the income shock will both lead to an increase in child mortality and worsened stunting. Because more children will be stunted when they reach school age, it is also likely that this will decrease educational outcomes. Income shock The income shock (​Δy / y​) used in the simulations comes from the World Bank Global Economic Prospects.98 The values come from the difference between projected GDP per capita growth for 2020 utilized in the Macro Poverty Outlook from the World Bank Annual Meetings of 2019 (pre-COVID-19), and the GDP per capita growth projections made in June 2020. Stunting The effect of the income shock on stunting is: ∂ Stunting ​ΔlnHCI = − ​γstunting ​ ​​_ ∂y ​Δy (1)​​ where ​​γ​ stunti​_​ng​​ = 10.2 × 0.034 = 0.35​as discussed in Appendix A. Although a direct value of ​​ ____ _ ∂ Stunting ∂ y ​is not avail- able, this is replaced with an elasticity from Ruel et al. (2013): ∂ Stunting _ y ​ ∂ y ​_ ​Stunting ​ = − 0.6 (2)​​ Inserting (2) into (1) yields the following expression for the direct effect of an income-induced increase in stunting on the HCI: ∂ Stunting y Δy ​ΔlnHCI = − ​γstunting ​ ​​_ ∂y ​_ ​Stunting ​_ ​ y ​Stunting (3) ​ For countries missing stunting data, the average rate for its income group is applied. 97 Because the parameter ​γ​stunting​​embodies the best alternative of the link between stunting to adult height and from adult height to earnings, the index can be expressed by relying just on stunting as a proxy for health. 98 World Bank (2020). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 81 Education The effect of the income shock on education of a child born today is expected to come through the effect on stunting: ​​(​ ∂∂Stunt ____ ∂ HTS ____ ​​ΔlnHCI = __ ∂ y ​HTS + ​ ∂ Stunt​​ ∂ y ​EYS)​Δy ϕ ____ EYS ____ ​ 625 ​∂​ Stunt ∂ Stunt ​(​4)​​ ∂ EYS ∂ HTS where ​ϕ = 0.08​ and ​​ _____ _____ ∂ Stunt ​ = − 1.594 years of education,​ and ​​ ∂ Stunt ​ = − 0.625 SD.​ Inserting (2) into (4) gives us the effect of the income shock on education, operating through increased stunting: 625 ( ​​ ∂ y ​​Stunting ​EYS)​​_ ϕ EYS _ _ ∂ Stunting y HTS _ _ ∂ Stunting y Δy ​ΔlnHCI = ​_ ​∂∂Stunt ​_ ​∂∂Stunt ​​ ∂ y ​​Stunting ​HTS + _ y ​Stunting (5)​​ Mortality The negative income shock increases child mortality, with the following effect on the HCI: ∂ U5MR ∂ lnsurvival _ ​ΔlnHCI = ​_ ∂ U5MR ​​ ∂ y ​Δy (6) ​ ∂ lnsurvival ∂ U5MR where ________ −1 ​​  ∂ U5MR ​is equal to ________ ​​ 1 − U5MR ​. Although a direct value of ______ ​​  ∂ y ​is not available, this is replaced with a semielasticity from Ma et al. (2020):99 ​∂ U5MR _ ∂ y ​y = − 0.013 (7) ​ Inserting (7) into (6) yields: Δy ∂ U5MR _ ∂ lnsurvival _ ​ΔlnHCI = ​_ ∂ U5MR ​​ ∂ y ​y ​ y ​ (8)​​ An additional shock to mortality is assumed to come from the change in access to health services mea- sured in months of disrupted access: ∂ U5MR ∂ lnsurvival _ ​ΔlnHCI = ​_ ∂ U5MR ​​∂ access ​Δaccess (9)​​ ∂ U5MR Although a direct value of ​​ ______ ∂ access ​is not available, this is replaced with a monthly access change to the elas- ticity of the under-5 mortality rate from Roberton et al.100 ​∂∂ U5MR _ _access access ​​U5MR ​ = 0.136 (10)​​ This value suggests a monthly relative increase in under-5 mortality of 13.6 percent given a one-month lack of access. Since the values that enter the index are annual, this is extrapolated to the year and inserted into (9): ∂ U5MR _ access _ ​∂ lnsurvival ​ΔlnHCI = _ _ Δaccess ∂ U5MR ​​∂ access ​​U5MR ​​access ​U5MR (11)​​ Under the baseline scenario, a change in the access to care of three months is assumed, thus in (11), access is equal to 12 months and ​Δaccess​is equal to 3 months. 99 The elasticities are disaggregated by income groups: for high income countries it is assumed to be 0; for upper middle income equal to -0.003; for lower middle income equal to -0.01; for lower income equal to -0.013. Ma et al. (2020) express these values as a 1 percent decrease in GDP per capita being associated with an increase of 0.13 under-5 deaths per 1000 children. Or an increase of 0.013 percentage points in under 5 mortality rates. 100 Roberton et al. (2020) offer three scenarios where the effect ranges from 8 to 34.5 percent. 82 Accum ulati on Interrupted? C OVID -1 9 and Hu m an Capita l HCI Putting all these pieces together gives us the total in HCI change due to the pandemic: ∂ U5MR ​​∂ access ​​U5MR ​​ access ​U5MR + ​ ∂ U5MR ​​ ∂ y ​​U5MR ​​ y ​U5MR + ​625 ( ∂ U5MR _ ∂ lnsurvival _ access _ y _ ∂ U5MR _ ∂ lnsurvival _ Δy ϕ _ EYS _ ∂ Stunting y ​ΔlnHCI = ​_ Δaccess _ _ ​ ​∂∂Stunt ​​ ∂ y ​_ ​Stunting ​ ​​ ∂ y ​​Stunting ​EYS)​​_ ∂ Stunting HTS _ _ y Δy ∂ Stunting y Δy ​∂∂Stunt HTS + _ y ​Stunting − ​γstunting ​ ​​_ ∂y ​_ ​Stunting ​_ ​ y ​Stunting ​ Note how the first component, the one related to access, enters independently from the income shock. 3.5 ANNEX 3B: COVID-19 SHOCK TO SCHOOL AGE COHORTS The shock to children who are presently in school is derived as in Azevedo et al. (2020), but using the data for the 2020 HCI. The shock to children operates through two channels: (a) the income channel, leading to increased dropouts, and (b) the school closure channel, leading to loss in learning and in school years. Recall that learning-adjusted years of school (LAYS) at pre-COVID-19 baseline (0) is: HT ​S​​ ​LAY ​S0​​ = EY ​S0​​× ​_ 625 ​ 0 The changes in income, how well governments can deliver education while schools are closed, and how long schools are closed are all expected to decrease LAYS. The number of out-of-school children is assumed to increase due to the income shock. These changes are calculated for each welfare quintile using data from 130 household surveys using the latest available Global Monitoring Database (GMD), separately for children ages 4 to 11 and 12 to 17. The shock from the GDP per capita growth projections is used to arrive at a new welfare value. This is achieved by assuming the shock is uniform across the distribution; thus the shape of the distribution is maintained, it is just shifted to the left. The shift of the household welfare of children moves children across welfare quintiles; the quintile thresholds are the same as those from the original welfare distribution. Finally, the quintile’s share of out-of-school children is used to get the new total number of out-of-school children.101 In essence, when household income drops, children move down the welfare quintiles (because the thresholds are maintained). With more children in lower welfare quintiles with higher shares of out-of-school children, there will be an overall increase in the share of out-of-school children, since the denominator (total number of children in the specific school age bracket) stays the same. The first component of the change in LAYS is the share of students who drop out due to the income shock (D). When children go to school, they experience in-person learning, which is assumed to be the most effi- cient learning mode. With school closures, children will experience different, less efficient, learning. The length of school closures differs according to scenarios that range between 3 and 7 months (Table 3.4). The effectiveness of different remote learning strategies deployed, and the scenarios, are linked to the country’s income group (see Table 3.5). 101 For countries without a household survey, the overall change in out-of-school rates for the country’s income group is used. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 83 Table 3.4: Length of school closure by scenario Months Optimistic 3 Intermediatev 5 Pessimistic 7 Table 3.5: Mitigation effectiveness by scenario and income group Low income Lower-middle income Upper-middle income High income Optimistic 0.2 0.28 0.4 0.6 Intermediate 0.1 0.14 0.2 0.3 Pessimistic 0.05 0.07 0.1 0.15 The second component of the change in LAYS is the share of the school year that is lost due to the closure and to the alternative learning modality (S): ​S = (​ 1 − mitigation)​× closur​eshare ​ of school year​ Harmonized test scores are assumed to change over a school year by a certain amount (​p​); the amount is dependent on the country’s income group (see Table 3.6). The learning of these children is compromised due to the closures and the limited effectiveness of the deployed learning modality. Table 3.6: School productivity (HTS points gained per school year) Points High income 50 Upper middle 40 Lower middle 30 Low income 20 The final component to the change in LAYS is the amount of learning that takes place under the remote learning scenario (H): ​H = S × p​ The change in LAYS is then: ​ΔLAYS = LAY ​S1​​− LAY ​S0​​ HT ​S​​ ​ΔLAYS = LAY ​S1​​− EY ​S0​​× ​_ 625 ​ 0 where ​LAY ​S​1​​ is equal to: ( ) ​(​HT ​S0​​− H​)​ ​LAY ​S1​​ = ​ ​EY ​S0​​− S − D​ ​​_ 625 ​ 4 UTILIZING Human Capital T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 85 T he World Bank’s Human Capital Index This chapter introduces two Utilization-Adjusted (HCI) captures the size of the income gains Human Capital Indices (UHCIs) that, as their when today’s better-educated and health- name suggests, adjust the HCI for labor-market ier children become tomorrow’s more productive underutilization of human capital.1 The UHCIs workers. Specifically, a child born today can expect are is designed to complement the main HCI, and to be HCI×100% as productive as a future worker as not to replace it. In part, this is because these two she would be if she enjoyed complete education measures have different purposes: the HCI is an and full health. But this implicitly assumes that index of the supply of a factor of production (in the when today’s child becomes a future worker, she future), whereas the UHCIs are a hybrid between will be able to find a job—which may not be the an index of factor supply (capturing investment in case in countries with low employment rates. human capital), and a utilization index (capturing Moreover, even if today’s child is able to find how efficiently that human capital is used in pro- employment in the future, she may not be in a duction). Moreover, there are numerous challenges job where she can fully use her skills and cognitive in defining and measuring utilization in a consis- abilities to increase her productivity. In these cases, tent way across diverse country contexts. As such, human capital can be considered underutilized, the UHCIs should be viewed as a first attempt to because it is not being used to increase productiv- address utilization in a simple way consistently ity to the extent it could be. For example, unem- across countries, and should be applied with cau- ployed future workers may be underutilizing their tion in policy analysis. human capital, as are those out of the labor force. Likewise, engineers driving taxis are underutiliz- Importantly, the HCI and UHCI only mea- ing their human capital because, even though they sure the effect of human capital on labor mar- are employed, they do not hold jobs in which their ket earnings and future GDP per capita. But this education increases their productivity. is only one benefit of human capital. In many other domains, human capital improves wellbe- In addition, a gender gap—which is not apparent ing and economic development. More–educated in the human capital dimensions cap­tured by the parents have children with better human capital HCI—emerges and deepens during the working outcomes, and women with more human capi- years. In many countries, women face worse jobs tal are more empowered. Even outside the cate- and income opportunities compared to men, even gories of “better employment” considered below, with the same human capital. As such, simply con- human capital can still increase productivity—for sidering the HCI by sex may give a partial view in example, smallholder farmers might use fertilizer terms of realizing the potential of human capital more efficiently—but just the increase is less dra- investments. matic than for other employment types. As such, 1 Prepared by Steven Pennings (spennings@worldbank.org) with helpful comments from Roberta Gatti, Aart Kraay, Michael Weber, Kathleen Beegle, Paul Corral, and David Weil, as well as from other internal, and external reviewers. See Pennings (2020) for an in-depth treatment. 86 U ti lizing Hu man Capita l incomplete utilization should not be interpreted to these investments being realized in terms of as there being no gains from human capital invest- income opportunities for women. ments, but rather that private labor market gains are smaller than they could be. 4.1 METHODOLOGY AND THE BASIC UHCI The two UHCIs take different approaches to mea- MEASURE suring utilization. In the basic UHCI, utilization is measured as the fraction of the working-age Both the basic and full UHCIs have a simple form, population that is employed. While this mea- as the utilization rate multiplied by the HCI: sure is simple and intuitive, it is not able to cap- (​ 1)​ UHCI = Utilization Rate × HCI​​ ture underutilization resulting from a mismatch between the skills and cognitive abilities required For the basic UHCI, this multiplicative form stems to do a job, and the skills and cognitive abilities of from its connection to economic growth. In the the people employed to do it. The full UHCI mea- long run, GDP is proportional to the number of sure adjusts for this mismatch by introducing the workers (employment) multiplied by the pro- concept of “better employment,” which includes ductivity boost that each worker gets from their the types of jobs that are common in high-pro- human capital.2 The basic UHCI inherits this mul- ductivity countries. tiplicative form, where the HCI captures the pro- ductivity boost from human capital, and the utili- Despite different methodologies, the basic and full zation rate captures employment.3 measures produce broadly similar utilization rates. Utilization rates are U-shaped in per capita income The HCI is derived to measure the effect of human across countries, first declining with income at lower capital on future GDP per capita so that pro- income levels and then rising at higher income jected future per capita GDP will be approximately levels. This feature of utilization rates implies that 1/​HCI​times higher in a “complete education and full UHCIs are low in the poorest countries where the health” scenario than in a ‘status quo’ scenario.4 This HCI is also low on average, but remain low over a definition implicitly assumes that utilization rates of wider range of lower middle-income countries human capital—such as employment prospects—are where rising HCIs are offset by declining utilization. the same in the ‘complete education and full health’ scenario as in the status quo scenario. Moreover, both UHCIs reveal starkly different gender gaps from those using the HCI. Girls have a Both UHCI measures are derived in a similar way, slight advantage over boys in human capital early in keeping with the economic interpretation of the in life, resulting in a higher HCI for girls on aver- HCI. However, for the UHCIs, utilization rates are age. But female utilization rates are typically lower now different in the status quo and full human cap- than those for males, resulting in lower UHCIs. ital scenarios. Specifically, both UHCI indices are While gender gaps in human capital in childhood derived as future GDP per capita under the status and adolescence (especially education) have closed quo relative to future GDP per capita with complete in the last two decades, large challenges remain health, education, and full utilization (Equation 2). 2 See box 4.1 for a derivation. More specifically, this requires a Cobb-Douglas production function, and the assumption that the cap- ital-to-output ratio is constant in the long run (one of Kaldor’s facts). 3 In the full UHCI, the utilization rate is defined as UHCI/HCI—and so satisfies Equation (1) by construction—but still turns out to have an intuitive interpretation (see Section 4.3). 4 Kraay (2018). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 87 This means that, in the long run, GDP per capita A natural concern is that in countries with low basic will be 1/UHCI times higher in a world of full uti- utilization rates, human capital, as measured by lization, full health, and complete education than the HCI, will have less effect on economic growth. in the status quo. 5 However, this is not the case.7 In the framework of the basic UHCI, an increase in human capital alone has (2) ​UHCI = ___________________________________ future GDP per capita (​ ​status quo​)​          ​ GDP per capita (​ ​full utilization, full health, complete education​)​​ future the exact same effect on long run economic growth as in the HCI, but it’s just that countries can do better by For the basic UHCI, the utilization rate is simply also increasing utilization (it is not one or the other).8 the employment rate of the working-age popu- lation. This is current employment ​L​, relative to The main data source for the basic utilization mea- a measure of potential employment under “full sure is “Employment-to-population by sex and age utilization” ​L​ ​,​the maximum theoretical employ- ⋆ (%) – Annual,” Age (Youth, Adults): 15-64, from the ment. The standard definition of the potential International Labour Organization (ILO), using the labor force is the working-age population 15–64 latest period available.9 The secondary data source years old. This definition is also adopted for ​L​ ​. ⋆ 6 is the World Bank’s Global Jobs Indicators ( JOIN) Employment ​L​is defined as the number of people Database, which has employment based on the 15–64 years old that are in paid employment (or same population age group, with a sample skewed are self-employed) to be consistent with the defi- towards low- and middle-income countries. Data nition of the potential labor force. is generally taken from the most recent source if both are available.10 The median year of the data As mentioned above, the basic UHCI takes the is 2017, with 95 percent of countries having data simple multiplicative form in Equation (1) because, from the 2010s. The basic utilization measure is in a standard production function, long-run GDP available for 185 countries. The measurement of per capita is proportional to human capital per the full UHCI is discussed in Section 4.3. worker ​h​ multiplied by employment L per capita (see Box 4.1 for a derivation). Future GDP per capita under the status quo (in the numerator of Equation 4.2 THE BASIC UTILIZATION-ADJUSTED 2) is proportional to ​hL​, and future GDP per capita HCI IN THE DATA under complete utilization and complete human capital is proportional to ​​h​​⋆​ ​L​​⋆​​ (​​h​​⋆​​ is complete Basic utilization rates are not strongly correlated human capital per worker) in the denominator with the HCI (correlation coefficient of 0.45), which of Equation (2). Because the proportionality means that countries’ UHCI scores will differ from factors are the same, and so cancel out, this can those in the HCI (Figure 4.1). Employment rates be rearranged as ​HCI = h / ​h​ ​​ multiplied by the basic ⋆ average around 0.6, which suggests that the UHCI ​ til ​(basic)​ = L / ​L​⋆​​. utilization rate U will, on average, be around 60 percent of the value 5 Just like the HCI, the UHCI can also be interpreted in terms of productivity: a child born today can expect to be only UHCI×100% as productive as she would be, on average, if she enjoyed complete education and full health, and her future labor was fully utilized. 6 Naturally, no countries will have employment rates of 1. But this is consistent with the approach in the HCI, where no country has perfect test scores or 14 expected years of schooling. 7 In the full UHCI, discussed in Sections 4.3-4.4, countries with very low “better employment” ratios will have GDP that is less sensi- tive to increases in human capital. But even there, improvements in human capital will still increase growth. 8 Technically, this is because the implicit assumption in the HCI is that basic utilization rates are constant across status quo and full human capital scenarios. A full employment assumption is not required. 9 Downloaded from https://www.ilo.org/shinyapps/bulkexplorer7/ on 13 December 2019. 10 In some cases, the more recent data source of employment data is not used if it is missing data for the full UHCI. 88 U ti lizing Hu man Capita l Box 4.1: Deriving the basic UHCI Future GDP per capita of the next generation (​y)​ in the status quo world is given by a Cobb- Douglas production function: ​y = A ​K​1−β​(hL)​β​/ N​ (1) where h represents human capital per worker under current policies, and L represents the number of workers under status quo employment rates. A is TFP, K is the amount of physical capital, and N is the future population. TFP and the future population are assumed to grow at the same trends in all scenarios. In an alternative world, there is complete human capital per worker, denoted ​h​⋆​, and com- plete employment of potential workers, denoted ​L​⋆​​. Long-run GDP per capita in this com- plete human capital-complete employment world is denoted ​​y​​⋆​​. As in Kraay (2018),_ _the production function can be rearranged so that the physical capital-to-output ratio ​K ​/ ​Y ​ is constant in the long run. Then, future GDP per capita under the status quo relative to the complete human capital, complete utilization scenario is given by: _ _ _ y ​A​1/β​_ ​(K ​ ​/_​Y )​ ​​(​1−β​)​/β​h L / N ​​y​⋆​​ = _________________    ​​A   ​1/β​​(K ​ ​/ ​Y )​ ​​(​1−β​)​/β​​h​⋆​​L​⋆​/ N ​ (2) ​= ​_L ⋆ _ ​L​ ​​× ​​h​ ​​ h ⋆ ​= Utilization (​ ​basic​)​× HCI​​ ​= UHCI ​(​basic​)​ of the HCI (Figure 4.2), though this varies substan- are likely because most people are so poor that they tially across countries. need to work outside the home to survive. At around 0.55, lower-middle-income countries have the low- Employment rates (basic utilization) are approx- est utilization rates, mostly because slightly higher imately U-shaped in log per capita income incomes make it feasible for people (especially (Figure 4.3).11 High-income countries have the women) not to work outside the home (see Section highest utilization rates (around 0.7 for the group 4.7 for a discussion of utilization rates by gender). as a whole). This is unsurprising, as it is difficult to have high per capita incomes with few people work- Employment rates vary widely among low- and ing. Low-income countries have utilization rates middle-income countries. While many low-in- of around 0.6 on average, though many low-in- come countries have high employment rates come countries also have extremely high employ- of around 0.8, others have employment rates of ment rates of around 0.8—like Madagascar (MDG), around 0.4—including Malawi (MWI), Nepal Burundi (BDI), and Mozambique (MOZ). High (NPL), and Afghanistan (AFG). In part, this may employment rates among low-income countries reflect the 2013 change in the ILO definition of 11 For lower-income countries, the U-shape is mostly driven by several outliers with extremely high utilization rates. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 89 Figure 4.1: Basic Utilization (Employment/Population) and HCI 0.9 QAT MDG SLB ISL TZA KHM VNM 0.8 BDI ZWE ARE CHE MAC Employment−to−Population SWE NZLNLD JPN MOZ DEU LBR DNKNOR CZE EST GBR UGA THA SYC Ratio (basic utilization) KWT BHR BLR LTU AUT AUS CAN SGP MLT LVA SVN FIN 0.7 CMR PRY KNAPER PLW OMN RUS HUN USA ISR CYPPRT IRL HKG URYBGRKAZ CHN ECU SVK LUX POL PAN IDN COL KOR TTO CAF NER MLI BEN MMR HND NIC ROU MYS FRA BEL CODBFA GMB TLSVUT DOM MUS CHL ESP GTM SLV AZE ARG MEX 0.6 GINETH COG BTN NRU PHL KEN BRAGEO KGZMNG CRI BRN SRB HRV ITA BWA BGD FJI JAM UKR ALB GHA LKA SLE CIV TGOHTI MNE GRC FSM SAU TUV GUY TON MKD ARM TUR 0.5 TCD NGA PAK PNGNAM IND RWALSO GAB TJK MDA BIH SSD AFG SEN ZAF MARTUN LBN SDN COM IRN 0.4 SWZ MRT MWI LAO EGY KIR DZAWSM UZB IRQ AGO NPL MHL PSE YEM ZMB 0.3 JORXKX 0.2 .3 .4 .5 .6 .7 .8 .9 Human Capital Index (HCI) ILO Data JOIN Data Figure 4.2: Basic UHCI versus HCI 0.9 0.8 Human Capital Index (UHCI) Basic Utilization−adjusted 0.7 ISL SGP JPN MAC SWE 0.6 NLD CHENZL CAN GBR EST DEUNOR DNK AUS FIN QAT VNM CZE HKG AUT SVN IRL ARE PRT KOR CYP 0.5 RUS BLR LTU MLTISRPOL LVA USA FRA BEL SYCBHR HUN THA CHN LUX ESP SVK OMN PER KAZ HRV ITA BGR CHL 0.4 KHM KWTKNAURY PLWCOL ECUMYS TTOMUS CRI BRN SRB ZWE PRY ROU MEX ARG GRC SLB IDN AZEKGZ UKR MNGALB MDG GEO MNETUR TZA NIC SLV PAN BRA KEN LKA DOM 0.3 BDI UGA TLS VUT HND MMR PHLJAM SAU NRU BTN FJI GTM MKD ARM MOZ CMR FSMTON BGD GUY MDA GMB BENCOG GHA BIH LBR BFA BWA HTI TGO TUV IND IRN UZB CODETH GIN TJK TUN WSM MAR LBN 0.2 CAF MLI NER SLECIV PAKPNG LSOSEN ZAF GAB EGY NAM KIR NPL DZA PSE NGARWA MWI LAO AFG COM XKX JOR TCD SDN MRTIRQMHL SWZ SSD AGO ZMB 0.1 YEM .1 .2 .3 .4 .5 .6 .7 .8 .9 Human Capital Index (HCI) Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Based on 169 countries with data available. Nicaragua and Nepal are highlighted in red. For Figure 4.2: Dashed line is 45 degree diagonal, and solid line is a fitted model where y=-0.14+0.88x employment to exclude own-use production These measurement issues also motivate using a workers (mostly subsistence agriculture), which more specific definition of employment in the full has been applied in some countries but not others.12 UHCI (discussed in the next section). 12 In Malawi, a household survey in 2005 reported employment rates of 0.8, twice the most recent figure (from 2017). Likewise, in Nepal a household survey in 2008 reported employment rates of 0.84, also around twice the most recent figure (also from 2017). The difference is likely due to exclusion of own-use production workers in 2017, though this is not well documented. 90 U ti lizing Hu man Capita l Figure 4.3: Employment-to-Population (basic utilization) and GNIPC 0.9 QAT MDG SLB ISL TZA KHM ZWE VNM 0.8 BDI ARE CHE SWE MAC NZLNLD JPN MOZ BHS GBR DEU DNK LBR Employment−to−Population UGA THA SYC CZE EST NOR BLR KWT CAN AUS AUT LTU BHR LVA MLT FIN SGP Ratio (basic utilization) BOL 0.7 CMR PRYPER RUS CHN HUN KAZ KNA BRB OMN PLW URY SVN PRT CYP ISR HKG USA IRL ECU BGR POL SVK PANTTO KOR LUX GNB IDN COL MYS NER CAF MLI MMR HND NIC ROU FRA BEL BEN TLS VUT CODBFAGMB DOM MUS CHL ESP AZE SLVBLZ GTM MEX VEN ARG KEN 0.6 ETH GIN KGZ COG BTN GEO MNGPHL SRB SUR CRIHRV BRANRU CUW BRN ITA BGD UKR JAM ALB FJI BWAMDV GHA LKA TGO HTI SLE CIV MNE GRC FSM TUV SAU CPV ARM GUY TONMKD TUR 0.5 TCD PAK NGA PNG IND NAM RWA TJK LSO MDA BIH GAB SSD AFG SEN MARTUN ZAF LBN COM SDN EGY IRN 0.4 MWI MRT LAO SWZ LBY UZB KIR DZA WSM IRQ NPL AGO MHL PSE YEMZMB STP 0.3 XKX JOR DJI 0.2 5 6 7 8 9 10 11 12 log GNI per capita (Altas) ILO Data JOIN Data Missing HCI data Figure 4.4: Basic UHCI and GNIPC 0.7 SGP ISL JPN SWE MAC 0.6 NZL NLD CAN CHE EST GBR FIN AUS DEU NOR VNM CZE HKGDNK QAT Human Capital Index (UHCI SVN Basic Utilization−adjusted PRT KOR AREAUTIRL BLR LTU CYP 0.5 POLLVA MLT ISR FRA BEL USA RUS HUN SYC BHR ESP LUX THA CHN SVK KAZ HRV OMN ITA PER BGR KHM KNA KWT 0.4 COL ECU SRB MYS CHL MUS URY PLW TTO CRI BRN ZWE PRY ROU MEX ARG GRC KGZ SLB UKR MNG ALB IDNAZE GEO MNE MDG KEN SLV LKA BRA TUR PAN TZA NIC HND DOM 0.3 BDI MMR TLS BTN VUT PHL ARM JAM FJI MKD NRU SAU MOZ UGA CMR GTM TON GMB BGD MDA FSM BIH BEN COG GHA UZB GUY BFA LBR HTI IRN TGO ETH TJK IND TUV BWA COD GIN MAR TUNWSM LBN 0.2 NER MLI CIV PAK PNGEGY DZA NAM GAB SLE KIRPSE CAF NPL LSO SEN ZAF AFG RWA NGA LAO MWI COM XKX MRT SDN SWZJORIRQ MHL SSD TCD ZMB AGO YEM 0.1 6 7 8 9 10 11 log GNI per capita (Altas) Source: World Bank calculations based on the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Based on 182 countries with data available. Working-age population is 15-64. To understand the implications of differences around 0.65, which is above the median, as against between the UHCI and HCI for long-run eco- that in Nepal, 0.37, which is around the 5th percen- nomic growth, consider the example of two coun- tile. This means that the basic UHCI score of Nepal tries, Nicaragua (NIC) and Nepal (NPL). These (0.18) is much lower than that of Nicaragua (0.33) countries have similar scores for the HCI (0.5), but (Figure 4.2, dark red dots). As mentioned above, very different employment rates (Figure 4.1, dark the increase in long-run per capita income mov- red dots). The employment rate in Nicaragua is ing to full human capital is 1/HCI times that in the T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 91 status quo, and long-run per capita income mov- traders selling household goods or food, where the ing to full human capital and complete utilization is majority of time is spent waiting for customers. 1/UHCI that in the status quo. An HCI score of 0.5 for both countries implies that long-run per capita While there is scope for human capital to increase incomes would double moving to full human cap- productivity in these jobs, it is more limited. Filmer ital. However, moving to full human capital and and Fox (2014) compare the income of household full utilization of that human capital results in long- enterprise owners of different education levels in run GDP per capita that is 3 times the status quo in four African countries. On average, the increase in Nicaragua, but 5.4 times the status quo in Nepal.13 income due to education, while positive, is much less than would be predicted given the number of The basic UHCI is fairly flat over a wide range years of schooling.14 Most developing countries suf- of log income before increasing (Figure 4.4). fer from high rates of mismatch between the level Specifically, the UHCI is almost flat moving from of education required for a job, and the education low income (0.23) to lower-middle income (0.26), of the people doing it: the well-known anecdote as higher HCI scores are largely offset by lower uti- of unemployed engineers driving taxis.15 In the lization rates. But then the UHCI increases rapidly literature, this is often referred to as “over-edu- to upper-middle income (0.32) and high income cation”, though a more appropriate description is (0.51), as both human capital and utilization rates “underutilization” as it is the lack of jobs and not increase together. the level of education that is the cause of the mis- match. In some regions, especially the Middle East and North Africa, underutilization is often associ- 4.3 THE FULL UTILIZATION-ADJUSTED HCI ated with self-employment, for example, while queuing for a formal sector job.16 One conceptual issue with the basic utilization measure (employment rate) is that it assumes that To address this, the full UHCI introduces a concept all jobs are the same in terms of their ability to uti- of better employment, which is designed to capture lize human capital. But in practice, a large share the employment categories where people can bet- of employment in developing countries is in jobs ter use their human capital (subject to available where workers cannot fully use their human capital. data). More specifically, better employment is defined For example, in the poorest countries, around half as non-agricultural employees, plus employers. of all workers work on family farms or as agricultural This definition is not intended as a value judgment, day laborers, where productivity is low (Merotto et but rather is based on the types of jobs that are rel- al. 2018). For the rest, around two-thirds of non-ag- atively rare in low-income countries, but are com- ricultural workers are self-employed or unpaid in mon in high-income countries—suggesting they family businesses. These include many small-scale are associated with higher productivity. The share 13 For the average low-income country, long-run incomes in the complete utilization and human capital scenario would be around 4.5 times those of the status quo (1/0.22≈4.5), compared with 2.5 times that with complete human capital alone (1/0.4≈2.5). 14 On average, those with a complete secondary education were only earning 60 percent more than those with no education, which is the equivalent of less than six years with an 8 percent return to education. Omitted variables such as parental income and ability mean that the six years is likely an overstatement. 15 See Battu and Bender (2020) for a survey. Another cause of the mismatch can be poor education quality, where those with a qualifi- cation aren’t able to perform the functions required. In this case, the cause of the engineer driving a taxi is because he or she is not able to perform the tasks of an engineer due to poor quality education. Handel et al. (2016) find that in 12 low- and middle-income countries, the overeducation/underutilization rate is 36 percent. Overeducation/underutilization rates vary across countries and can depend on how it is measured. 16 Handel et al. (2016) and Gatti et al. (2013). 92 U ti lizing Hu man Capita l of employment in better jobs increases from about closer to being fully utilized. This means the uti- 20 percent in low-income countries to 80 percent lization scores of countries with higher levels of in high-income countries (Merotto et al. 2018). The human capital should be more heavily penalized main categories excluded from the definition are by a lack of better employment.19 subsistence own-account/family agriculture, small- scale traders, and landless agricultural laborers, Putting these concerns together suggests that the as these employment types are only common in full UHCI should depend on the better employ- low-income countries—suggesting they are more ment rate (BER) (as a share of the working-age likely to have lower productivity. By using a more population), rather than the raw employment rate. specific definition of employment, the full UHCI However, the full utilization rate is not simply the also avoids variation in utilization rates caused by BER, because this fails to adjust for how much differences in the definition of employment that human capital there is to underutilize if people are affected the basic UHCI. not in jobs where they can fully use the human cap- ital. Instead, utilization rates for those without bet- The definition of better employment is based on ter jobs should depend inversely on the HCI (relative the way that the work is organized, rather than to a natural minimum). The full utilization measure whether the job is formal or informal. For exam- captures both concerns. The full UHCI is a weighted ple, non-agricultural employees could be formal or average of the country’s HCI (for those in better informal.17 Better employment involves work orga- employment), and the minimum HCI (for the rest) nized in a team consisting of at least an employer and is described further in Box 4.2 (and Equations and an employee, where employees are paid for 3 and 4). The full UHCI can also be derived based their work (rather than out of familial obligation). on the increase in long-run GDP per capita moving This allows a minimum degree of specialization from the status quo to a world with full human cap- and organization, which helps to boost productiv- ital and full utilization in better employment (see ity and allows for people to use their skills.18 Pennings, 2020).20 A second conceptual issue with the basic measure is In terms of data, the BER is constructed as the that utilization should be relative to potential, which employment rate (as in basic utilization measure), will depend on how much human capital there multiplied by the share of employment in better is to underutilize. That is, a doctor working as an jobs (SEBJ). The measurement of the SEBJ requires agricultural laborer results in severely underuti- data on the number of employers, non-agricul- lized human capital, whereas the human capital of tural employees, and total employment. The pri- a worker with no education doing the same job is mary source is the ILO series “Employment by 17 The definition of formal employment varies across countries, but it generally refers to the coverage of the worker with respect to benefits like unemployment insurance, pensions, sick leave or annual leave. 18 Better employment differs from “decent work” (ILO) and “good jobs for development” (World Bank 2012). 19 A final technical issue is that some of the increase in GDP in the basic UHCI comes from utilizing people’s time, rather than utilizing their human capital. The full UHCI also addresses this concern (see Pennings, 2020). 20 It is important to acknowledge that the definition of better employment and the full UHCI are stylized for simplicity and cross-country data availability. In reality, many people without better jobs can partially use their human capital to increase produc- tivity beyond that of raw labor. For example, education is positively correlated with high-yield-variety seed choice among Indian farmers (Foster and Rosenzweig, 2010). Moreover, healthier people may be more productive laborers. Assuming that only educa- tional human capital (not health human capital) is underutilized outside better employment results in the same U-shape pattern in per capita GNI, but with higher utilization rates (and UHCI) for low-income countries (not reported). Moreover, there are many examples of people without better jobs using their human capital to its full extent, such as self-employed professionals. However, the availability of cross-country employment data limits the amount of nuance possible in this regard. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 93 Box 4.2: Definition of the full UHCI The full utilization measure is a weighted average of the utilization rates of those in better employment, and the utilization rate of the rest of the working-age population. Workers in better employment are assumed to be as productive as their human capital allows— their human capital is fully utilized (utilization rate of 1). All others, a fraction (1- BER) of the working-age population, are assumed to be only as productive as “raw” labor, and hence any excess human capital is underutilized. In the HCI, “raw labor” has productivity of ​HC​Imin ​ ​ = 0.2.​This is the productivity of a worker with zero years of schooling, and the worst possible health outcomes.a In contrast, the potential productivity of a worker if they were in better employment is just ​HCI​. Hence the worker’s productivity relative to potential, or utilization rate, is ​HC​Imin ​ ​/ HCI​. For example, in a country with HCI=0.4, workers without better employment will be half as productive as they could be if they were in better employment (0.2/0.4), and so their utilization rate is 0.5. This means that a shortage of better employment leads to more severe underutilization in coun- tries with more human capital. The full utilization measure is given by: ​ tilization (​ full measure)​ = BER × 1 + ​(1 − BER)​× _ HC​Imin ​ ​ (3) U ​HCI ​ The full UHCI is the full utilization measure multiplied by the HCI, as in Equation (1). This means that the full UHCI is a weighted average of the HCI (for the share of the population in better employment) and the minimum HCI (for the rest of the working-age population): (4) ​ HCI ​(full measure)​ = BER × HCI + ​(1 − BER)​× HC​Imin U ​ ​ a The minimum HCI score is derived by assuming zero years of schooling, complete stunting, and zero chance of adults ​ ​ = 1 × ​e​0.08×​(​0−14​)​​× ​e​​(​0.65×​(0−1)​+0.35×​(0−1)​)​/2​ ≈ 0.2​. See Kraay (2018), equations 9-12. The probability of surviving to age 60. ​HC​Imin survival to age 5 is assumed to be 1, as this doesn’t affect the growth calculations. sex, status in employment, and economic activity data on total agricultural employment (which is (Thousands),” using the most recent year available. 21 more widely available). The full utilization measure The secondary source is the JOIN database. At the is available for 161 countries. time of writing, the public JOIN dataset provides a split by status in employment or economic activity, not both, so the SEBJ is calculated using an unpub- 4.4 FULL UTILIZATION-ADJUSTED HCI IN lished version constructed from the underlying THE DATA microdata. The most recent data source is used if both the ILO data and JOIN are available. For many The full utilization measure has the same U countries, there are missing data on the number of shape in log per capita income as the basic utili- agricultural employees. In these cases, the number zation measure, and similar mean values overall of agricultural employees is interpolated using ILO (0.62) and for each income level, though with less 21 Downloaded from the ILOSTAT website on 20 February 2020 (defined using ICSE-93). 94 U ti lizing Hu man Capita l Figure 4.5: Full Utilization Rate and GNIPC QAT 0.9 ARE KWT ISL MAC 0.8 BHR DEUSWE CHE DNK NOR OMN BLR SYC EST JPNAUT RUS KNA MLT PLW NZL CAN AUS HUNLVA LTU GBRHKG FIN BGR CZE ISR SGP SVN BRN 0.7 CAF KAZ SVKPRT TTO SAU CYP FRA NLD BEL IRL LUX TCD BWA HRV POL Full Utilization Rate ETH ZAF MUS NRU ESP LBR MLI ROU ARG MYS URY KOR NER SWZ PRY PAN UKR PERMNE KHM MKD DOMCHNBRA CHL ITA 0.6 MOZ COD UGA RWA YEM MRTCMR TJK PAK HND BTNTUN AGO GTM SLV NAM GUY PHLMHL TUV JAM FJI SRB MEXCRI SLE CIV SLB COM BIH MDG BFA GINTZA ZMBCOG NIC EGY TUR GRC BDI GMB GHA IDN GEO DZA THA GAB AFG BEN SENBGD ARM AZE COL MMR TLSVNM MNG UZBVUTFSM WSM LKA TON ECU HTI 0.5 TGO NPL KGZ ZWE LSO MDA LAO MARPSE JOR XKX IRN IND ALB LBN KEN 0.4 0.3 0.2 5 6 7 8 9 10 11 12 log GNI per capita (Altas) ILO Data JOIN Data ILO (interpolated agric.) Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Based on 161 countries with data available. dispersion (Figure 4.5). However, the U-shaped The full UHCI also has the same shape in per cap- pattern has quite different causes from those driv- ita income as the basic UHCI (and similar mean ing the basic utilization measure. For the full mea- values for each income level, Figure 4.6). However, sure, the highest-income countries have around for the lowest-income countries, the UHCI value 70 percent of the working-age population in bet- converges almost exactly to 0.2, with little varia- ter employment, (Figure 4.9) which drives the tion (as against wide variation in the basic UHCI). high utilization rate. For low-income countries, The reason is that 0.2 is the minimum HCI score, only around 10 percent of the working-age popu- which is the assumed productivity of “raw labor” lation are in better employment, so the utilization for those without better employment. rate for these countries is mostly determined by how much human capital there is to underutilize. In the ten lowest-income countries, the HCI​≈​1/3, 4.5 COMPARING THE UTILIZATION so ​HC​I​min​/ HCI​ is around 0.2/0.33=0.6—close to the MEASURES full utilization rate for those countries in Figure 4.5. The full utilization rate falls from low-income While the full and basic utilization measures have to middle-income countries, as higher rates of the same U-shaped relationship with per capita human capital mean that there is more human income, they often differ substantially for indi- capital to underutilize (and the BER only increases vidual countries (Figure 4.7, correlation of only slightly). 0.6).22 The strongest correlation is for high income 22 This is driven by a number of countries on the left side of Figure 4.7 in MENA and elsewhere where a high fraction of total employ- ment is classified as better employment (such as wage employment), and a number of countries, often in EAP, with lower rates of wage employment on the right side of Figure 4.7. Some of these EAP countries are also penalized in the full measure by having a high HCI that increases the potential to underutilize human capital. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 95 Figure 4.6: Full UHCI and GNIPC 0.7 SGP MAC SWE JPN ISL 0.6 EST HKGDNK DEU CAN NOR CHE NZLFIN QAT GBR AREAUTAUS Human Capital Index (UHCI SVN IRL NLD CZE PRT CYP Full Utilization−adjusted BLR MLT ISR FRA LVA LTU BHR KOR BEL 0.5 RUSPOL HUN HRVSYC ESP LUX OMNSVK KWT BGR ITA BRN KAZ PLWKNA MUS TTO 0.4 UKR SRB MNE PER CHN CHL GRC MYS ARG URY SAU ROU CRI VNM MEXTUR MKDTHA BRA NRU BIH PRY UZB MNG SLV COL PAN ARM AZEJAM ECUDOM LKA GEO PHL 0.3 TJK KHM KGZ NIC HND BTN IDN MDA TUN PSEDZAGUYFJI ALB WSM XKX JOR IRN EGY TON ZAF BWA GTM FSM TUV NAM ETHNPL MMR BGDSLBGHA MAR MHL GAB KEN GMB UGARWA HTI ZWE COMCOG CMR SEN YEMZMB IND LAO VUT SWZ PAK TLS LBN MDG TZAMRT CIV AGO BDI COD MOZ TGO AFG BFA BEN 0.2 NERCAF SLE LBR GIN TCDMLI LSO 0.1 6 7 8 9 10 11 log GNI per capita (Altas) Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Based on 161 countries with data available. countries, because in order to generate high per occur mostly for countries with a low HCI, which capita incomes, employment rates need to be high mechanically shrinks any differences in utilization and those people working need to be productive. rates when forming the UHCI.23 But for lower income countries, the drivers of high utilization vary across measures, and the similar- ity of average scores is coincidental. Specifically, 4.6 DISAGGREGATION BY REGION employment rates (basic utilization) are often high in low-income countries because people cannot Regions line up similarly according to the UHCI afford not to work, though with a lot of variation and to the HCI (Figure 4.10), though UHCI scores due to the inconsistent cross-country classifica- are lower. Sub-Saharan Africa (SSA) has the low- tion of work in subsistence agriculture. In con- est HCI (of around 0.4), and also the lowest UHCIs trast, there is little variation in full utilization rates (of around 0.23). South Asia has similar UHCI, but across low-income countries, because low-income higher HCI (reflecting slightly lower utilization countries have little human capital to underutilize. rates). Latin America and the Caribbean (LAC) and the Middle East and North Africa (MENA) are next, For the UHCI, the scores of individual coun- with HCI scores around 0.56 and UHCI scores tries are very similar in the full and basic mea- around 0.35, though MENA does relatively better sures (Figure 4.8, correlation 0.93). In part this for the full UHCI than the basic UHCI, reflecting is because full and basic UHCI have the HCI as higher rates of wage employment. East Asia and a common component. It is also because the dif- the Pacific (EAP) are marginally higher, followed ferences between the two utilization measures by Europe and Central Asia (ECA). 23 The one exception is Vietnam, which has high employment rates, but a low fraction of that is in better jobs. These differences re- main prominent in the full UHCI because of Vietnam’s high HCI score. 96 U ti lizing Hu man Capita l Figure 4.7: Basic Utilization versus Full Utilization QAT 0.9 ARE KWT ISL MAC 0.8 BHR DEUSWE CHE DNK NOR OMN EST JPN BLRSYC NZL AUT PLW KNA RUS MLTAUS BGR HKG ISR LTUCAN HUN LVA FIN SGP GBR CZE SVN 0.7 SAU BRN CAF SVK FRALUX BEL TTO CYP IRLPRT KAZ NLD TCD BWA HRV Full Utilization ETHMUS NRU ESP POL ZAF ARG MLI ROU MYSKOR URY LBR SWZ MNE UKR NERPAN PER PRY MKD ITA BRACHLDOM CHN KHM 0.6 YEM AGO MHL MRT TUN TJK RWA NAM GUY PAK TUV PHL FJISRB JAM BTN MEX CRI SLVCOD HND GTM CMR UGA MOZ SLB COM BIH EGY CIV SLE GRC NIC ZMB TUR COG GIN GMB IDN TZA MDG DZA GAB GHA GEO BFA THA BDI SEN ARM BGD AZE AFG FSM LKA MNG BEN COL VUT VNM JOR NPL WSM UZB MDA TON HTI TLSMMRECU 0.5 XKX PSE LAO IRN MAR LSO TGO ZWE IND ALB KGZ LBN KEN 0.4 0.3 .3 .4 .5 .6 .7 .8 .9 Basic Utilization High income Upper middle income Lower middle income Low income Figure 4.8: Basic UHCI versus Full UHCI 0.7 MACSGP SWE JPNISL 0.6 HKGNOR DEU DNK QAT CANCHE EST FINGBRNZL AUT AUS ARE IRL SVN NLD CZE ISR FRA CYPPRT BLR MLT BHRBELLVA LTU KOR 0.5 ESP RUS POL HUN HRV LUXSYC OMN SVK KWT BRN KNA ITA BGR Full UHCI PLW KAZ MUS TTO 0.4 SAU MNEUKR ARG SRB CHL GRC MYS URYPER CHN ROU CRI TUR MEX VNM MKDNRU BRA THA UZBBIH PAN SLV PRY COL MNG ARM JAM LKA GEOAZE 0.3 TUN PSE DZA XKX ZAF WSM MDA FJI TJKIRNGUY PHL DOM ALB IDN NIC KGZ ECU KHM JOR EGY BWA TONBTN HND GTM NAM TUV FSM MHL NPL GAB MARETH GHABGD MMRKEN SLB SWZCOM PAKLBNIND COG VUT TLS ZWE YEM RWA LAO SEN CIV HTI GMBCMR UGA AGOMRTAFG SLE ZMB TGO CODBFABEN MOZ BDI TZAMDG 0.2 TCD CAFMLI GINLBR LSONER 0.1 .1 .2 .3 .4 .5 .6 .7 Basic UHCI High income Upper middle income Lower middle income Low income Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Based on 161 countries with data available. Dashed line is 45-degree diagonal. For Figure 4.7: fitted line is y=0.37+0.43x. For Figure 4.8: fitted line is y=0.05+0.89x 4.7 DISAGGREGATION BY GENDER typically lower than those for males (Figure 4.11 and Figure 4.12, Figure 4.13 and Figure 4.14) leading the Many of the trends above are driven by differences in UHCI also to be lower for females than males (Figure utilization rates by gender. While the HCI is roughly 4.15 and Figure 4.16, Figure 4.17 and Figure 4.18). Male equal across genders with a slight advantage for and female UHCI increase proportionately, but females relative to males, female utilization rates are with a constant gap for females (implying a larger T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 97 Figure 4.9: Better Employment Rates (BER) and GNIPC 0.9 QAT 0.8 ARE ISL MAC 0.7 BHR KWT DEUSWE CHE DNK NOR (share of Working−age pop) EST JPN BLR OMN NZLAUT RUS SYC MLT CAN GBR AUS HKG SGP LVA LTU CZESVN FIN HUN ISR 0.6 KNA PLW PRTCYP Better Employment BGR NLDIRL SVK BRB BRN FRA BEL LUX KAZ POL HRV TTO SAU ESP KOR 0.5 MUS MYS UKR ROU ARGURY ITA SUR PERMNECHNNRUCHL CUW SRB 0.4 PRY MKD BLZ MEXCRI PAN GRC BRA MDV BWA ZAF DOM TURVEN KHM SLV BIH PHL JAM VNM GUYFJITHA 0.3 ETH TJK HND NIC CPVTUN MNG UZBBTN IDNAZE COL GEO ARM LKA EGY DZA NAM GTM ECU MDA WSMTUV BOLPSEJOR SWZALB XKXMHLIRN KGZ FSMTON 0.2 RWA NPL MMRPAK COG DJI BGDSLB GHA MAR GAB GMB UGA YEMCOM CMRSTP ZWE TLS VUT MRT SEN ZMB HTI TZA KEN CIV IND LAO AGO LBN 0.1 BDI MDGCOD MOZ AFGLBR TGO BEN CAFSLEBFA MLIGIN NER TCD 0 LSO 5 6 7 8 9 10 11 12 log GNI per capita (Altas) ILO Data JOIN Data ILO (interpolated agric.) Missing HCI data Figure 4.10: Regional Average UHCI or HCI 0.8 0.69 0.59 0.57 0.55 0.6 0.48 0.47 0.45 0.40 UHCI or HCI 0.38 0.38 0.38 0.35 0.34 0.4 0.32 0.26 0.24 0.24 0.23 0.2 0 SSA SA LAC MENA EAP ECA HCI Basic UHCI Full UHCI Source: World Bank calculations based on the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Figure 4.9 based on 171 countries with data available. Figure 4.10 reports regional averages. percentage gap at low UHCI scores). The gender gap generally, female employment rates are strongly is larger for the basic measure than the full measure. U-shaped in the level of income, whereas male Surprisingly, when women join the labor force, they employment rates are much flatter.25 The largest more rapidly move into better employment. More 24 gaps in utilization rates across genders (for both 24 See Pennings (2020), Figure 25A. 25 See Goldin (1995). Klasen (2019) shows that the U-shaped pattern of female employment rates is mostly due to region fixed effects, and not the development path for an individual country. 98 U ti lizing Hu man Capita l Figure 4.11: Female Employment-to-Population Figure 4.12: Male Employment-to-Population (basic utilization) and GNIPC (basic utilization) and GNIPC 1 1 QAT ARE NER MDG TZA KHM SLB OMN BHR KWT ISL ZWE VNM GTM MDG SLB JPN CHE ISL HND NIC PRY CZE MLT BHSNZLNLD MAC 0.8 BDI TZA KHM VNM 0.8 MLI BGD MMRPAK BOLIDN LKA SLVBLZFJI THA ECU COL PER CHNMEX PAN KNA SAU EST DEU GBR SGP SWE AUS DNK NOR ZWE BDI MOZ UGA LBR DOMKAZMYS MUS RUSCRIHUNPLW AUT HKG CAN TTO SVN KOR USA Employment−to−Population MOZ SWE CHE CMR SYC Employment−to−Population GMB ETH KGZ BLR MDV POLURY VEN IRL SYC BFAGNB IND BTNPHL NRU ROU ARG SVK LVA LTU BRB PRTCYP FIN LBR BLR LTUEST NZLNLDDNK MAC DEU NOR BEN AZE BGR CHL ISR CAN BRATUR Ratio (basic utilization) UGA BHSJPN FIN GBR TLS SUR LUX Ratio (basic utilization) LVA AUS VUT TUV BRBCZE SVN PRT AUT CAF AFG TJK MNG MAR IRQ GEOJAM IRN ITA FRA ESP BRN BEL NER CMR THA RUS ISR SGP USA COD HTI FSM TUN GUYSRB BWA TON LBN HRV GRC TGO BOL BGR CYP IRL LUX HKG TCD GIN CIV KEN COG UKR EGY DZAALB CUW HUNSVK LSO SDN 0.6 CAF COD GNB GIN KENTLS PER KAZ PRY CHN POL URY KNA MLT PLW FRA BEL QAT 0.6 YEM MRT GHA ARM MKD MNE GAB BEN COG VUT KOR ESP SEN CPV SLE GHA GEO ECU COL ROUHRVTTO PAN SLE RWA COM NGA PSE BIH LBY BFA GMB MLI MMR UKR MNG IDNAZE SRB CHL CUW BRN MYS ARG NPL WSM BWABRA ARE UZB ETH HND BTN ALB DOM VEN ITA TGO PNGKIRAGO NAM MHL ZAF NIC PNG JAM SLV SUR MNE MUS KWT MWI JOR CPV CRI SSD MDA SSD HTI KGZ CIV NGAMDA PHL NAM BLZ ARM MEXNRU GRC SWZ XKX BHR LAO 0.4 RWA TCD FSM MKD GUY GTM TON FJI MDV 0.4 ZMB DJI LKA ZAFGAB SWZ STP BGD LAO BIH TUV MWI LSO SEN KIRAGO TUR COM UZB WSMLBY OMN ZMB STP MRT NPLTJK MARTUNMHL 0.2 AFG PAKIND SDN LBN SAU 0.2 EGY DJI DZA IRN XKX PSE JORIRQ YEM 0 0 5 6 7 8 9 10 11 12 5 6 7 8 9 10 11 12 log GNI per capita (Altas) log GNI per capita (Altas) Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Data for 182 countries. Working age population is 15-64. Figure 4.13: Female – Full Utilization Rate and Figure 4.14: Male – Full Utilization Rate and GNIPC GNICPC 1 1 QAT ARE BHR KWT OMN NRU SAU MAC ISL DEU ISL CHE 0.8 SWE MAC 0.8 KNA MLT DNK NOR SWE AUT HKG RUS HUN EST DEU DNK NOR NZL AUS SGP Full Utilization Rate EST CHE BLR CZE PLW LVA BRN CAN Full Utilization Rate LTU BGR GBR ISR LVA FIN AUT SVK SVN LTU FIN RUS CAN NZL GBR AUS KAZMYS CYP NLD LUX BGR KNA ISR QAT ETH BWA ARG PRT KOR FRA BEL IRL HUN CZEPRT SVN MLT CYP HKGSGP IRL LBR EGY ZAF PER MEX PRY ROUHRV POL PAN ESP TCD KAZ SVK PLW FRANLD TCDMLI YEM PAK TUN BRN BEL LUX SLV UKR BTNAGODZA FJI MNE MKD CHNTUR URY RWA MRT HND PHL GUY GTM MHL CHL ITA NER MLI BWA HRV POL ESPKWTARE NER BIH DOM CRI LBRETH URY COGSLB IDNJOR ZAF NRU COD BGD NIC COM PSE JAM SRB GRC 0.6 UKR AGO PRY MNE ROU DOM PER ARG PAN BHR KOR 0.6 BDI MDGSLE GMB TZA CMR CIV GHA SEN XKX AZETUV GEO ARM GAB THA IRN ITA TLSVNM COD GIN SRB MKD JAM MYS CHL CHN AFG HTIGIN MMR BFA BEN UZB FSMWSM VUT TON COL ECU AFG RWA TZA SLE CIV YEMCMR PHL GUY CRI MNG MDG BFA MRT SLBHND GTM MHLFJI COM NIC MDA PAK GHA BTNSLV GEO THA MEX ARMBIHCOL GRC TGO KGZ LAO IND MDA ALB BDI GMBBEN TGO COG SEN MNG VUTIDNAZE GAB MMRBGD VNM UZB LAO TUN FSM OMN HTI TLS WSM TON ECU KGZ EGY ALB TUR SAU IND DZA 0.4 JOR 0.4 PSE IRN 0.2 0.2 5 6 7 8 9 10 11 12 5 6 7 8 9 10 11 12 log GNI per capita (Altas) log GNI per capita (Altas) Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Based on 138 countries (Figure 4.13) or 141 Countries (Figure 4.14) with available data. measures) are for several oil/gas producers: Bahrain, Figures 4.19 and 4.20 break down the HCI and Kuwait, Oman, Qatar and Saudi Arabia. These UHCIs by gender and region. In almost all regions, countries have very high male employment rates— the female HCI is higher than male HCI (equal almost all of which is in wage employment (perhaps for South Asia). However, the opposite is true for due to migrant workers)—but low or average female the UHCI: in almost all regions the female UHCI utilization rates. is lower (the only exception is ECA for full UHCI). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 99 Figure 4.15: Female – Basic UHCI and GNIPC Figure 4.16: Male – Basic UHCI and GNIPC 0.7 0.7 SGP ISL NZL NLD ISL MAC CHE SWE GBR SWE ARE 0.6 MAC 0.6 CZE KOR HKG CAN NLDSGP FIN EST DEUAUS IRLQAT EST NZL CAN CHE NOR DNK NOR VNM VNM AUT FIN DEU DNK GBR POL SVN PRTMLT BHRCYP BLR LTU SVN AUS LVA CZE HKG USA PRT AUT FRA BEL 0.5 Human Capital Index (UHCI) 0.5 ISR IRL BLR CHN OMN HUN ISR Human Capital Index (UHCI) Basic Utilization−adjusted CYP RUS LVA ESPITA Basic Utilization−adjusted RUS POL FRA THA LTU LUX KOR BEL USA PER COL MEX SVK KWT ECU KAZ MYS CRI CHL KNA TUR MLT HRV HUN LUX PRY SAU KGZ SRB BGR ARGPLWGRC URY THA SVK ESP IDNAZE HRV 0.4 SLV ROU BRN 0.4 BGR KAZ CHN QAT NIC HND MNG ALB IRN GTM PAN PER URYKNA FJI MNE PLW MMR BGD UKR DOM SLB UKR SRB ITA IND SLB PHL GEO JAM NRU CHL BRN ARE ROU MDG BTN DZA MDG MNG GEOALB ECU COL MYS TZA PAK MARFSM TUN TON ARM MKD BDI TZA ARG GRC BIH GUY 0.3 IDN AZE PRY MNE CRI BHR 0.3 GMB BEN CMR TLS UZBVUT EGYPSE TUV MEX PAN KWT BDI NER HTI BFA ETH AFG TGO KGZ TLS MDA VUT SLV ARMJAM WSMIRQ CMR GHA MMR DOM COG GHA MDA JOR GAB NIC MLI CIV XKX BWA COG HND BTN PHL MKD NRU COD LBR GIN GMBBEN LBR BWA YEMSEN SDN COD MRT GIN SLEBFA HTI FSM TONBIH TUR 0.2 TGO COM PNGKIR 0.2 NER ETH PNG UZB GUYFJI RWA SLE TCD NGA LAO MHL ZAF GTM AGO KIR MLI CIV BGD NGA LAO ZAFGAB TUV OMN RWA WSM SEN COM MARTUNAGO 0.1 TCD MRT IND MHL IRN SAU 0.1 PAK EGY SDN DZA AFG PSEXKX JOR IRQ YEM 0 0 6 7 8 9 10 11 6 7 8 9 10 11 log GNI per capita (Altas) log GNI per capita (Altas) Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database Notes: Based on 148 Countries with data available. Figure 4.17: Female – Full UHCI and GNIPC Figure 4.18: Male – Full UHCI and GNIPC 0.7 0.7 SGP MAC SWE MAC ISL SGP ARE HKGSWE NOR 0.6 EST FIN DNK 0.6 DEU CAN QAT CHE CAN HKG DEU NZL ISL DNK NOR NZL CHE EST GBRAUTAUS GBR AUS BHR LTU SVN IRL FIN NLDIRL LVA PRT AUT NLD CZE SVN KOR CZE ISR CYP PRT MLT Human Capital Index (UHCI) FRA FRA Human Capital Index (UHCI) RUS CYP BEL POL OMN ISRBEL Full Utilization−adjusted 0.5 POL MLT HUN Full Utilization−adjusted HUN 0.5 BLR RUS LVA ESP KWT LUX HRV SVK KOR ESP HRV LTU QAT LUX SVK SAU ITA KNA BRN BGR KAZ ARE BGR KAZ CHNMYS KNA BRN ITA TUR UKR PLW PER MNE CHLPLW GRC ARG 0.4 SRB MNE URY BHR 0.4 SRB MEX NRU CRI KWT UKR ROU URY CHL VNM CHNROU ARG GRC MYS BIH PRY MKD VNM PER CRI UZB SLV PSE THA TUNDZA AZE IRN COL MEX NRU JOR ECU PAN MNGGEO MKD THA EGYIDN MNG XKX PHL ARM FJI JAMPRYCOLDOM MDA PAN GEOALBJAM GUY 0.3 UZB ARM BIH SLVAZE TUR OMN 0.3 KGZ HND NIC BTN WSM GTM TON DOM PHL ALB GUY FJIECU BWA MDA KGZ ZAF BGD FSM ZAF NIC PAK GHA BWA IDN WSM SAU ETH MMR MHL GAB HND TUN TON BTN FSM GTM GMBHTI YEM COMCOG TLS SLB IND VUT AGO TUV ETH RWA MRT CMR MMR DZAMHL IRN GAB SEN LAO GHA BGD SLB LAO VUT EGYPSEJOR BDI COD AFG MDG TGO BEN TZA CIV GMB RWA SENCOG TLS CMR COM LBR GIN MDG TGO HTI TZA SLEBFA MLI PAK IND 0.2 BDI NER COD SLEBFA LBR AFG BEN MRT CIV GIN YEM TCD MLI AGO 0.2 NER TCD 0.1 0.1 6 7 8 9 10 11 6 7 8 9 10 11 log GNI per capita (Altas) log GNI per capita (Altas) Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. Notes: Based on 138 Countries (Figure 4.17) or 141 Countries (Figure 4.18) with available data. The largest gender gaps for the basic UHCI are including social norms.26 However, the full UHCI in MENA and South Asia. In these two regions has smaller gender gaps, in part because of small the female basic UHCI is very low, reflecting low gaps in how much human capital there is to employment rates driven by a variety of factors, underutilize. 26 More specifically, female labor force participation rates in MENA are low for women without tertiary education, whereas those rates are much higher for women with tertiary education (Gatti et al 2013). This may reflect high reservation wages, and because tertiary education is required for public-sector jobs that are perceived to be safer for women. 100 THE U TILIZAT IO N -ADJU ST E D H U M AN CAP I TA L I N DEX Figure 4.19: Gender Gaps by Region 0.6 0.57 0.54 0.45 0.45 0.41 0.39 0.4 0.38 0.35 0.33 UHCI or HCI 0.32 0.29 0.27 0.25 0.24 0.22 0.22 0.20 0.2 0.13 0 Latin America & Caribbean South Asia Sub−Saharan Africa HCI − Male HCI − Female Basic UHCI − Male Basic UHCI − Female Full UHCI − Male Full UHCI − Female Figure 4.20: Gender Gaps by Region 0.8 0.71 0.67 0.60 0.59 0.6 0.57 0.55 0.48 0.47 0.47 0.44 UHCI or HCI 0.43 0.41 0.41 0.39 0.4 0.36 0.33 0.33 0.19 0.2 0 Europe & Central Asia East Asia & Pacific Middle East & North Africa HCI − Male HCI − Female Basic UHCI − Male Basic UHCI − Female Full UHCI − Male Full UHCI − Female Source: World Bank calculations based on the 2020 update of the Human Capital Index for HCI data, the World Development Indicators, the International Labour Organization and the Global Jobs Indicators Database. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 101 Box 4.3: Closing gender gaps in human capital outcomes: Where do we go from here? The HCI approach implicitly assumes that human capital investments translate into produc- tivity through labor market opportunities. However, there is considerable variation in how human capital is utilized in terms of paid work and labor markets. In particular, considerable and well-documented gaps exist in labor market opportunities between men and women. Globally, only 1 in 2 women participate in the paid labor force, while 80 percent of men do. Across countries, the gender wage gap persists at around 20 percent, on average.a Women work in lower-paying occupations and jobs. Across the globe, only 1 in 5 firms have a female top manager.b While these outcomes might in part reflect optimizing decisions within the family (for example, see Chioda, 2016, for evidence from Latin America), evidence shows that a variety of constraints explain some portion of these gaps, ranging from the lack of child care and adequate leave policies to social norms that create barriers to women work- ing. These norms include those that put a disproportionate responsibility for domestic work and child care on women, as well as those that result in occupational sex segregation, sex- ual harassment, and mobility restrictions. Women must also contend with differential con- straints in access to finance and markets, a great divide in access to digital technology, and legal/regulatory barriers to start and grow firms.c All these factors result in wasted potential in terms of realizing economics gains from human capital investments in girls. Looking only at the sex-disaggregated HCI misses an important reality concerning gender gaps in how human capital is utilized. For human capital to translate into productivity, the human, who owns the capital, needs to be employed in work where they can use their human capital. In 2020, boys and girls growing up in Peru have the same HCI score of 0.6. However, only 62 percent of women in Peru are employed, compared to 78 percent of men, resulting in a basic UHCI that is 10 percentage points lower for females than for males.d Countries can act to enable women’s full participation in labor market opportunities. Provision of affordable child care options, parental leave policies, and flexible work options can accommodate women’s entry into formal work and help women and men redistribute and balance demands at home and at work.e Safe transport allows women to go to the workplace, while pay transparency can increase women’s power to negotiate equal pay for equal work. Improved access to digital technology for women can unlock potential gains from the digital era. These range from accessing online education to expanded income-gen- erating opportunities through flexible online gig work and e-commerce entrepreneurship.f Resources need to be mobilized to ensure that women and men have equal access to liveli- hoods and economic opportunities. Source: Prepared by Daniel Halim. a ILO (2018). b World Bank Enterprise Surveys. Retrieved from World Bank Gender Data Portal. c In low- and middle-income countries, only 54 percent of women have access to mobile internet, compared to 74 percent of men GSMA. (2020). The Mobile Gender Gap Report 2020. GSMA Connected Women. https://www.gsma. com/mobilefordevelopment/wp-content/uploads/2020/05/GSMA-The-Mobile-Gender-Gap-Report-2020.pdf d Pennings (2020). e Olivetti and Petrongolo (2017). f Dammert et al. (2014); World Bank (2016a); Alatas et al. (2019). 5 Informing policies to protect and build HUMAN CAPITAL T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 103 T he HCI 2020 update appears in a context 5.1 GOOD MEASUREMENT: NECESSITY, of urgent choices for policy makers in all NOT LUXURY countries. Strategic decisions made now have the power to protect and strengthen coun- As the COVID-19 crisis continues to unfold, good tries’ human capital, and with it their economic data and measurement are more vital than ever. future. Yet fiscal constraints and numerous competing priorities raise the risk that the need for urgent In addition to documenting pre-COVID-19 action delays the investments required in mea- changes in human capital across 174 countries, surement. In fact, measurement enables effective the HCI 2020 update has established a base- action. line for tracking the pandemic’s future effects on human capital. A further task for the update Better measurement and transparent informa- has been identifying pathways of influence from tion can be transformational in safeguarding COVID-19 to human capital outcomes in the and strengthening human capital. By generating short and longer terms. This may facilitate the a shared understanding among diverse actors, identification of entry points for policy. Using measurement can shine a light on constraints that the HCI methodology to quantify the gaps that limit progress in human capital. Through this pro- will likely emerge in health, skills, and knowl- cess, effective measurement can facilitate political edge because of COVID-19, this analysis under- consensus based on facts and muster support for scores the urgency of protecting and sustaining reforms. Measurement also enables policy makers the recovery of human capital, which will be a to target support to those who are most in need, cornerstone of countries’ postcrisis recovery and which is often where interventions yield the high- future economic growth. est payoffs. As policy implementation moves for- ward, measurement provides feedback to guide Good measurement and data are essential for course corrections. these policies to be well targeted and cost-effec- tive. But effective policies do not arise from thin If measurement can improve policy results around air. They are shaped and guided, and repeatedly human capital in ordinary times, its importance is undergo course corrections using the evidence multiplied during a crisis. Governments that can that reliable measurement can provide. access and use relevant data in real time are better able to act in a coordinated way on multiple fronts. To underscore this point, this report’s final pages In the case of COVID-19, they can monitor the evo- identify key ways measurement and data can con- lution of disease transmission and continuously tribute to policy success; map short-term and update control strategies, while responding to the longer-range agendas for strengthening the mea- immediate and long-term effects of the economic surement of human capital; and link them to sev- crisis on households and communities. Measuring eral specific policy changes necessary to protect how well children are growing, whether they are human capital in the wake of COVID-19. learning, and how financial stress and insecurity are 104 Inform ing p olicies to protect and bui l d hum an capita l affecting their development is a necessity, not a lux- accumulation, including across socio-economic ury. It is essential to design and target policies that groups and geography, and how policies can can remediate the pandemic’s negative impacts. affect it. Some key measurement improvements At a time when demand for government spend- can be achieved in the short term (for example ing is surging, and fiscal space is limited, data and on test scores, see Box 5.2). Longer-term efforts its transparent communication are vital to ensure will demand a more sustained commitment from accountability for how scarce resources are used. countries and development partners. The power of measurement to support transfor- 5.2.1 A short-term measurement agenda mative action in difficult situations extends beyond Due to the dramatic changes in household incomes public health emergencies. For example, data are and service delivery driven by COVID-19, there is especially important in countries affected by fra- an immediate need to measure the pandemic’s gility or conflict, though measurement is far more welfare impacts. However, social distancing is lim- difficult to carry out in these settings. Insecurity iting the way in which traditional surveys are col- and the lack of robust institutions hinder data col- lected by enumerators who visit families. Phone lection and, in turn, the ability of governments to surveys have helped answer this challenge by take action informed by evidence. Fortunately, helping reach households remotely and cheaply. innovative methods have recently enabled some Various research centers and institutions, includ- progress in understanding human capital dynam- ing the World Bank, have implemented phone ics in fragile contexts (Box 5.1). surveys in the wake of COVID-19.1 Emerging evi- dence from these surveys confirms that family incomes have dropped rapidly in many settings, 5.2 BEYOND THE HUMAN CAPITAL INDEX and that a large share of families have become food insecure. The HCI offers a bird’s-eye view of human capital across countries. By benchmarking the productiv- Phone surveys are relatively inexpensive, which is ity costs of shortfalls in health and education, the important at a moment when resources are espe- index has catalyzed new conversations within gov- cially scarce and countries face many compet- ernments, bringing discussion on human capital ing priorities. They are well suited for gathering accumulation to the level where decisions about information about behaviors (including access resource allocation are actually made. This is an to health services and uptake of remote learning important achievement. arrangements) or outcomes (such as income and consumption) subject to rapid variation. They are However, as a measurement tool, the HCI has sub- likely to return more reliable and informative data stantial limitations. For example, it does not speak when they build on existing information bases, to distributional or geographical differences within pointing to the importance of triangulating with countries. And while it focuses on what matters— existing initiatives. outcomes—it does not chart the specific pathways that each country needs to follow to accelerate Facility phone surveys are a complement to house- progress in human capital. Much greater depth hold phone surveys. These can document, for in measurement and research is needed to bet- example, how prepared health facilities are to man- ter understand the dynamics of human capital age COVID-19 patients and identify bottlenecks 1 Some of the emerging messages from these surveys are summarized in chapter 3 of this report. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 105 Box 5.1: Innovative data collection in fragile contexts: Examples from West Africa and the Middle East and North Africa Epidemics affect people’s health, and they also disrupt livelihoods and well-being through school closures, workers placed on furlough, restrictions on transportation and gatherings, and closing of international borders. As such, at the height of the Ebola epidemic in West Africa, in addition to assessing the impact of the disease on people’s health, it was also important to measure and monitor the epidemic’s socio-economic impact. However, given the nature of the epidemic, it was impossible and unethical to deploy enumerators to the field for data collection in face-to-face interviews at households and in communities.  In 2014, capitalizing on the proliferation of mobile phone networks, and building on the expe- riences of the mobile phone survey initiative called Listening to Africa (L2A), high-frequency mobile phone interventions were designed and implemented to provide rapid monitoring of the socio-economic impacts of the Ebola crisis in Liberia and Sierra Leone. Two nationally representative surveys, each conducted in Liberia and Sierra Leone when the crisis broke out, were used as the baseline for anchoring estimates in a representative dataset. In Liberia, researchers drew on the country’s Household Income and Expenditure Survey, which had to curtail fieldwork in August 2014. In Sierra Leone, they used the Labor Force Survey, which had completed fieldwork in July 2014. These existing surveys provided a database of phone numbers and household characteristics, which eventually became the sample frame for the phone survey. Data were then collected through call centers either nationally or internationally to reach over two thousand respondents in each country. Although phone surveys cannot replace face-to-face household surveys in all contexts, the experience in Liberia and Sierra Leone illustrated substantial benefits of such innovation in specific circumstances and for specific data collection needs, particularly the ability to col- lect timely data in volatile and high-risk environments.   Implementing surveys in a rapidly evolving context involves myriad challenges, including the lack of a relevant and reliable sample frame. For example, excluding displaced pop- ulations from national sample frames threatens the representativeness of socioeconomic surveys and consequently provides a skewed understanding of the country. As the size of forcibly displaced populations increases globally, it is urgent to devise strategies to include these populations in nationally representative surveys. The sampling procedure undertaken for the Syrian Refugee and Host Community Surveys (SRHCS), implemented in Lebanon, Jordan, and the Kurdistan region of Iraq over 2015-2016, offers valuable insights on over- coming survey-implementation challenges to obtain representative estimates in challeng- ing contexts. In the absence of updated national sample frames for host communities, and given the lack of comprehensive mapping of forcibly displaced populations, geospatial seg- menting was used to create enumeration areas where they did not exist. Data collected by humanitarian agencies, including the United Nations High Commissioner for Refugees (UNHCR) and the International Organization for Migration (IOM), were used to generate sample frames for displaced populations.   Source: Based on Hoogeveen and Pape (2020). 106 Inform ing p olicies to protect and bui l d hum an capita l Box 5.2: Leveraging National Assessments to Obtain Internationally Comparable Estimates of Education Quality The HCI highlights the need for regular and globally comparable measurement of learning to assess the quality of a country’s education system. Although most data on education quality included in the HCI currently come from assessments that are designed to be comparable across countries and over time using psychometric methods, they are often infrequent and do not yet cover all countries. Leveraging national learning assessments can help bridge the gaps in learning data. Most countries regularly conduct some form of assessment that can be augmented with short mod- ules of globally benchmarked and validated items to construct to construct globally compa- rable measures of education quality.a Though there is no comprehensive bank of globally benchmarked items, there are items from international assessments that can be incorporated into national assessments as “linking items”. These linking items provide commonality with international assessments, enabling learning outcomes to be placed on a global scale.b For instance, the 2021 National Assessment for Secondary Schools will enable Bangladesh to produce globally comparable learning outcomes. To allow comparison of national education quality on a global scale, the following countries have recently fielded or are planning to include linking items from international assessments in their national assessments. Sri Lanka. In 2009, the national assessment included linked TIMSS items. Subsequent national assessments in the country have maintained linking items with TIMSS to allow international com- parability. The score produced from the national assessment is used in the World Bank’s HCI. Uzbekistan. Before 2019, there was no internationally comparable learning outcomes data avail- able for Uzbekistan. The launch of the 2018 HCI galvanized the government toward measure- ment of education quality, and in 2019, with the support of the World Bank, the country con- ducted its first ever nationally representative and internationally comparable assessment (using TIMSS linking items) for grade 5 students in Mathematics, now part of the country’s 2020 HCI. Nigeria. Besides an Early Grade Reading Assessment conducted in 2014 for only four of the 37 states in the country, Nigeria’s learning data had been sparse until recently. The HCI 2018 empha- sized the need for a nationally representative and internationally comparable assessment of learn- ing outcomes in Nigeria. The Nigerian National Learning Assessment (NLA 2019), supported by the World Bank, is the first nationally representative learning assessment conducted in Nigeria using an internationally recognized methodology. The NLA 2019 measures student learning at grade 4 and grade 8 in core subjects of Mathematics, English, and Science and includes linking items to allow comparison on an international scale. Although not yet available, it will allow for inclusion in a future HCI of a nationallly representative and globally comparable learning measure for Nigeria. Relatively few linking items are currently available from international assessments, necessi- tating a cautious approach informed by country contexts: ensuring that the selected linking items align with the country’s national grade-level curriculum, are translated according to the protocols of the international assessment, are piloted in the country, and are not too easy or too difficult for the target population; that similar testing conditions are arranged as for inter- national assessments; and that a sufficient number of items are selected to provide reliable internationally comparable estimates of education quality in the country. a Birdsall, Bruns, and Madan (2016) and UNESCO (2018). b Kolen and Brennan (2004). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 107 in the delivery of routine health services, includ- establishing well-functioning vital registry systems ing immunization and maternal and child health to record such basic events as births and deaths services. Administrative data can also be used to is still a work in progress: less than 70 percent of monitor many aspects of service provision—at countries record such events, and progress to fill a low marginal cost, because these data already these gaps has been slow (Box 5.3). The quality exist in most countries. These existing datasets of school enrollment data, which in the index is could provide valuable insights, but they are often based on administrative records, is highly variable poorly linked, of varying quality, and inaccessible in low- and middle-income countries, particularly to groups outside of government. Similarly, big for the lower and upper secondary cycle. Finally, data can also be leveraged. For example, data from benchmarking learning outcomes internationally mobile phone records have been used to monitor has been a challenge, both across countries and mobility (which is important to modulate disease especially over time. This has significantly con- containment), to nudge behavior, and to improve strained the coverage of the long-run analysis of service delivery, including delivering educational changes in measured human capital. These chal- content. Digital technology and data can be har- lenges are heightened in fragile countries: in some nessed to provide social protection benefits more cases, data to inform various HCI components equitably and efficiently, both in the immediate simply do not exist; in others, data are too old and and in the longer run. likely do not sufficiently capture the rapid deteri- oration of human capital that can occur in fragile 5.2.2 Tackling long-term measurement needs contexts. In addition, comparable data, including In addition to solutions that can be deployed over time, for refugees, displaced persons, and rapidly, countries need strategies to improve the host populations are extremely limited. measurement of human capital in the longer run. An upcoming World Development Report will be The process of creating the 2020 HCI update has focused on data, and this report is not the place brought that message home. The process of data for a comprehensive description of the complex curation for this update has been a productive and rapidly evolving measurement landscape. opportunity to improve data quality jointly with However, it is worth touching on several areas counterparts. For example, thanks to close col- where better-funded and coordinated data collec- laboration with the Ministry of Human Resource tion and use could improve the understanding of Development in India, it was possible to signifi- human capital accumulation and effective inter- cantly improve upon publicly available data for ventions to accelerate it. school enrollment and arrive at a measure of expected years of school (EYS) constructed on the One such area concerns the long-term conse- basis of actual age-specific enrollment rates, which quences of interventions that have proved suc- capture enrollment more precisely. 2 cessful in the short run. For example, there is well-established evidence that conditional cash However, many gaps in the measurement of transfers (CCTs) have improved a variety of health internationally comparable key dimensions of and education outcomes within a few years of pro- human capital still persist (Box 5.2). For example, gram inception. However, there is relatively little 2 Expected years of school, conceptually, is just the sum of enrollment rates by age from age 4 to age 17. However, since age-specific enrollment rates are seldomly available, data on enrollment rates by level of school are used to approximate enrollment rates for the age bracket. In India, enrollment rates provided and used for EYS calculation are age specific and thus there is no need to ap- proximate the values. 108 Inform ing p olicies to protect and bui l d hum an capita l evidence on whether and how the increased time have textbooks? Are health centers stocked with spent in school under the CCT led to better learn- the necessary drugs? ing outcomes and improved labor market oppor- tunities. Projects should be prepared to monitor a Beyond assessing fundamental inputs, countries wide variety of potential outcomes, including edu- need answers on how to improve quality. These cational attainment, socioeconomic changes, and include understanding whether teachers actu- health indicators. Similarly, long-term evidence ally master the curriculum they are teaching and on the efficacy of some types of interventions if physicians diagnose diseases and treat them often relies on findings from small pilots that were appropriately. Selection mechanisms and incen- not followed by country wide scale-up, and ques- tives also matter for the quality of services. For tions therefore remain about the generalizability example, pay for performance has been widely of promising findings. introduced and requires evaluation. How can it best be managed and at what level? Private-sector Administrative data—for example, linking educa- financing and delivery also have the potential to tional assessments and hospital records to taxation improve service quality. But how can countries records, social security contributions, or health make sure that quality improves while services insurance via unique identifiers—allow such eval- remain affordable? Rapid advances in informa- uations. The benefits of user-friendly administra- tion and communications technology (ICT) like- tive data systems are vast, since they can inform wise hold promise to improve service delivery. policy choices about the design of cost-effective But reliable strategies to make this happen are not interventions, allow regular monitoring of key obvious and will differ across country contexts. outcomes, and support decision-making in real Additionally, quality reflects management capaci- time, all at low marginal cost. However, within ties and choices. What management interventions a country, administrative data are collected by improve service delivery in cost-effective ways? a variety of ministries and other entities, often And how can countries measure the quality of resulting in a patchwork of systems that does not management itself in the social sectors? favor integration and optimal use. Taking advan- tage of these data requires expertise that is scarce Administrative data can answer some of these in many countries. Finally, legitimate privacy con- questions but cannot provide insights into behav- cerns also restrict access and can make data linking iors and competencies. Surveys such as the Service incomplete or impossible. Delivery Indicators (SDI) can help. SDI are nation- ally representative facility surveys that measure A related question is understanding the “produc- the quality of services received by average citi- tion function” of health and education outcomes zens in primary health care centers and primary from the service delivery perspective. This issue schools.3 SDI collects data on critical inputs and is essential for designing effective interventions provider performance, and in the case of schools, and systems for quality health care and education. children’s learning. These types of data allow gov- It is even more pressing now in a post-COVID-19 ernments and service providers to identify gaps in world, where extensive remediation will be needed service provision, link financing inputs with health to compensate for the losses of human capital and education outcomes, and understand the caused by the shock. Unanswered questions are margins along which social sector spending fails to numerous and start at a basic level. Do students translate into quality services. SDIs are important 3 See https://www.sdindicators.org/. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 109 platforms for innovation and research, including 5.4 A DATA-DRIVEN HEALTH SECTOR measuring the quality of management in schools RESPONSE and hospitals. The immediate priority for countries fighting The analysis of delivery systems needs to advance COVID-19 remains containment and elimination in parallel with a deeper understanding of how of the novel coronavirus. Global efforts, such as human capital accumulates through the life improvements in testing and access to a safe and course. For example, evidence points to the nodal effective vaccine, will need to accompany local importance of early childhood years for lifelong measures to test, trace, and isolate carriers of cognitive, physical, and socioemotional develop- infection; to support the use of non-pharmaceuti- ment. However, systematic measurement of skills cal interventions such as masks and social distanc- in early years is a prerogative of very few coun- ing; and to implement targeted lockdowns when tries. Even when those measures are available, the necessary. Strengthening public health surveil- evolution of health status, cognitive abilities, and lance capacity will be essential to the timeliness non-cognitive skills during early childhood is not and effectiveness of these interventions. Robust well understood. Similarly, measuring skill—cog- surveillance requires the ability to collect, analyze, nitive and non-cognitive—among adolescents and and interpret relevant health-related data and use adults is still rare in most countries. these data to plan, implement, and evaluate con- trol actions. With most pandemics being of zoo- Advancing the long-term measurement agenda notic origin, closer coordination between health just described will require purposeful invest- and the agriculture sector will be instrumental to ments. In turn, funding measurement is a way to prevent future outbreaks, in keeping with a “one increase the efficiency and impact of future policy health” approach. action across multiple domains. By supporting the political economy of reform processes and guid- Complementing strong surveillance, it is essen- ing policy choices towards cost-effective solutions, tial to step up health services to care for COVID- better measurement and data use are investments 19 patients while maintaining the delivery of core that pay off. health services. COVID-19 highlights the need to invest in primary health care, with strong front- line delivery systems. In low- and middle-income 5.3 BUILDING, PROTECTING, AND countries, priority measures to strengthen pri- EMPLOYING HUMAN CAPITAL IN A POST mary health care may focus on reproductive and PANDEMIC WORLD child health and nutrition; infectious disease con- trol programs for HIV, tuberculosis, and malaria; Governments are now working under intense and community-based health promotion and dis- pressure to roll out policies across multiple sectors ease prevention. In middle- and higher-income in response to COVID-19. Measurement is essen- countries, a focus on improving healthy longev- tial to ensure that these policies are strategically ity, addressing non-communicable diseases, and designed and well implemented, and that they get linking primary care practitioners more tightly to results. What might effective policy solutions look disease surveillance networks will go a long way like, in the domains most important for human toward increasing resilience. In the face of wid- capital? While a companion paper to the HCI ening health disparities, it is essential to ensure 2020 update (World Bank, 2020b) discusses policy that disadvantaged households and communi- responses to COVID-19 in detail, below are some ties have access to quality and affordable care. In of the broad directions these responses may adopt. the past, disruptions of the health and economic 110 Inform ing p olicies to protect and bui l d hum an capita l Box 5.3: Data quality and freshness in the components of the Human Capital Index The HCI has proved a useful tool for policy dialogue, in large part because it incorporates human capital outcomes that are easily recognizable, consistently measured across the world, and salient to policy makers. While there are multiple aspects of human capital that can be measured in sophisticated ways, the relatively straightforward components of the HCI provide a snapshot of some of the most vital aspects of human capital accumulation. And while this section makes a case for better measurement of aspects of human capital outside the HCI, it is worth noting that even the fundamental components included in the index suffer from significant data gaps and quality issues. The components of child and adult survival used to compute the HCI are based on data on birth and death rates by age group. These data are primarily sourced from national vital reg- istries that are mandated to record the occurrences and characteristics of vital events like births and deaths. Vital statistics are essential to the measurement of demographic indica- tors like life expectancy and to identifying health priorities for the population. Vital statistics can also help target health interventions and monitor their progress. However, the coverage of vital registries varies widely; only 68 percent of countries register at least 90 percent of births that have occurred and only 55 percent of countries cover at least 90 percent of deaths (see panels A and B).a,b Birth registration has increased by only 7 percentage points (from 58 percent to 65 percent) in the past decade,c and in Sub-Saharan Africa, only eight countries have coverage of 80 percent or more for under-5 birth registration.d Stunting serves as an indicator for the prenatal, infant, and early childhood health environ- ments. The JME database that collates and reports global stunting data reports data for 152 countries, of which 33 have data that are more than five years old. In 10 countries, the most recent survey is over 10 years old. Gaps also remain in education data. The EYS measure is based on enrollment data that national governments provide to the UNESCO Institute for Statistics (UIS). Of the 174 coun- tries that form part of the HCI 2020 sample, 22 countries rely on primary enrollment data that come from 2015 or earlier. Since primary enrollment data are typically the most con- sistently reported, the issue of data freshness is of even greater concern for other levels of school. There are also significant gaps in time series data on enrollment rates. Of the 103 countries included in the 2010 HCI sample, 22 countries were missing primary enrollment rates for 2010. Data gaps are more numerous at other levels of schooling—over 30 coun- tries were missing secondary-level enrollment data for 2010, and 42 countries were missing these data at the pre-primary level. Finally, the latest update to the Global Dataset on Education Quality that produces the harmonized test scores covers 98.7 percent of the school-age population. However, of the 174 countries that have an HCI, 14 rely for test score data on Early Grade Reading Assessments (EGRAs) that are not representative at the national level. A total of 65 coun- tries (roughly 37 percent of the sample) rely on test score data that are from 2015 or earlier. There are also significant gaps in sex-disaggregated data across HCI components. The JME reports disaggregated stunting data for only 56 percent of the 887 surveys that are part of T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 111 the database. While sex-disaggregated enrollment rates are reasonably complete at the primary level, they are missing at the lower secondary level for 29 of the 174 countries that are part of the HCI 2020 sample. Sixteen countries in this sample are also missing disag- gregated test score data. As a result of these gaps, 21 of the 174 countries in the 2020 sam- ple do not have sex-disaggregated HCI scores. These gaps in disaggregated data span all regionse and income groups.f The credible and consistent measurement of human capital outcomes is essential to identi- fying priority areas for policy intervention, informing the design of those policies and track- ing their effectiveness over time. While high-quality data collection can doubtless be a costly undertaking, countries can also explore more cost-effective ways of monitoring the health and education outcomes of their citizens. For instance, instead of bearing the costs of partic- ipating in an international assessment like PISA or TIMSS, Uzbekistan incorporated assess- ment items into their national learning assessment that would allow for a linking with TIMSS (see Box 5.2). Panel A: Coverage of live births registration 90 percent or more 75-89 percent 50-74 percent Under 50 percent No data Source: © 2020 Mapbox © OpenStreetMap 112 Inform ing p olicies to protect and bui l d hum an capita l Panel B: Coverage of Death Registration 90 percent or more 75-89 percent 50-74 percent Under 50 percent Less than 90 percent 75 percent or more Between 70 and 79 perc.. Less than 75 percent No data © 2020 Mapbox © OpenStreetMap a https://unstats.un.org/unsd/demographic-social/crvs/#coverage; Boundaries and names shown and the designations used in this map do not imply official endorsement or accepatance by the United Nations or the World Bank b For countries that lack robust vital registries, the IGME (that reports on child mortality) and the UNPD (that reports on adult mortality) must rely on filling data gaps using population censuses, household surveys and sample registrations combined with model life tables. All these data must then be modeled to produce mortality rates. c UNICEF, A Passport to Protection: A Guide to Birth Registration Programming (New York: UNICEF, 2013). d United Nations Statistics Division, December 2014, “Coverage of Birth and Death Registration”. e Thirteen of the 21 countries missing a sex-disaggregated HCI score are from Sub-Saharan Africa; two each from East Asia and the Pacific, Latin America and the Caribbean and South Asia; and one each from the Middle East and North Africa and Europe and Central Asia. f Three of the 21 countries missing a sex-disaggregated HCI score are high-income countries, four are upper-middle- income, seven are lower-middle-income and six are low-income. status quo have sometimes enabled countries unexpected opportunity in many countries to introduce bold health-system reforms. 4 In to renew the commitment to universal health that sense, these difficult times may offer an coverage.5 4 See McDonnell, A., Urrutia A., and Samman (2019). 5 World Bank (2020c). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 113 5.5 PREVENTING LOSSES IN LEARNING 5.6 REINFORCING RESILIENCE AMONG VULNERABLE PEOPLE AND COMMUNITIES Along with more and better investments in health, a broad range of interventions are needed in other In the face of sharp declines in income, support sectors to get human capital accumulation back on to poor and vulnerable households is essential to track, both in the short and longer terms. Due to 6 mitigate the crisis impact and to sustain access to school closures and economic hardships, the cur- services and food security. In the first phase of the rent generation of students stands to lose signifi- pandemic response, the consensus on social assis- cantly in terms of learning and non-cognitive skills tance programs has been to cast the net wide, to now, and to lose earnings later in life. Strategies to avoid excluding any of those in need. In the medi- remediate schooling losses will require designing um-term, these interventions need to be reas- and implementing school re-opening protocols sessed and complemented or replaced by policy adapted to the specificities of the pandemic. At a measures geared toward an inclusive and sustain- minimum, these will involve protective equipment able economic recovery with support for employ- and supplies, health screening, and social distancing. ment and livelihoods (including with active labor Tailored teaching and learning resources, especially market policies that help match workers to new for disadvantaged children, are urgently needed in jobs and upgrade their skills), as well as assis- many settings to make up for lost learning. 7 tance to small and micro-enterprises.9 In parallel, strengthening social services, including counsel- Deeper reforms will need to follow to sustain ing, will help mitigate impacts on mental health access to schooling and promote children’s learn- and disruptions in people’s social networks. ing at all stages: starting from cognitive stimula- tion in the early years, to nurturing relevant skills COVID-19 has exacerbated many forms of in childhood and adolescence. Building blocks inequality, notably gender gaps. School closures for success will include better-prepared teachers, and a reduction in health services can interrupt better-managed schools, and incentives that are the trajectories of adolescent girls at a critical life aligned across the many stakeholders in educa- juncture. With women-owned firms primarily tion reform. The efforts that countries have made concentrated in informal or low-paying sectors, in providing continuity with remote learning the lack of basic formal social protection excludes during the pandemic could carry benefits beyond women and their families from buffers against the current emergency. Appropriately structured economic shocks, exactly at a time when they are online learning can facilitate the acquisition of being hit the hardest. Risks of gender-based vio- those competencies, such as collaboration and lence can also be heightened during times of cri- higher-order cognitive skills, that are increasingly sis, isolation, and confinement.10 These effects are essential in the changing world of work. To shape 8 amplified in fragile settings. resilient education systems, countries will need to draw lessons from this worldwide distance-learn- Deepening inequalities make targeting interven- ing experience and expand the infrastructure for tions to the most disadvantaged—and particularly online and remote learning. to children in their early years—an imperative, 6 World Bank (2020a) 7 World Bank (2020b). 8 Reimers and Schleicher (2020). 9 World Bank (2020d). 10 World Bank (2020d). 114 Inform ing p olicies to protect and bui l d hum an capita l to prevent setbacks that are likely to compromise accumulates strong human capital during her crit- lifetime health, education, and socioeconomic ical years of growth and development, it is because trajectories among the most vulnerable. These a large network of people and institutions have interventions should have an explicit gender angle contributed to the process. Parents decide what to to help progressively close the gaps that are now feed a child, when to take her to the doctor, and being magnified by COVID-19. whether and for how long to send her to school. Families make these choices within communities that transmit norms and may help households in 5.7 COORDINATING ACTION ACROSS need. In turn, communities rely on services that, SECTORS AND ADOPTING A WHOLE-OF- in many contexts, are provided largely by the pri- SOCIETY APPROACH vate sector, including non-governmental organi- zations. Finally, governments act to provide pub- COVID-19 has underscored the interdependence lic goods, address externalities, and ensure equity. that exists among multiple sectors that are fun- Wise public policy choices, informed by measure- damental for human capital accumulation. These ment, facilitate the shared achievement of human include health, education, infrastructure, water capital, and make it more than the sum of its parts. and sanitation, information technology, and oth- ers. Complex linkages connect these domains. For The COVID-19 crisis has put all the links in this example, proper hygiene contributes to limiting network under strain, not least the governments diffusion of the virus. In turn, reduced transmis- themselves. Under these conditions, progress sion is often a pre-condition to re-open schools. depends on leadership that recognizes the impor- Digital technologies enable educational continu- tance of building a future in which all children ity when physical re-opening cannot be achieved. can reach their potential. In the months and years But many poor and marginalized communities ahead, with limited fiscal space, protecting core lack access to digital tools. These links point to spending for human capital will challenge policy the need for ambitious infrastructure and other makers in many countries, regardless of their lev- investments in many countries to expand access to els of income. Yet, by making these investments water, sanitation, and digitalization as key enablers with a view to the future, countries can emerge of human capital accumulation. from the COVID-19 crisis prepared to do more than restore the human capital that has been lost. Connections across sectoral and social boundar- Ambitious policies informed by rigorous mea- ies emphasize the value of policy approaches that surement can take human capital beyond the lev- engage diverse stakeholders. Nurturing a nation’s els previously achieved, opening the way to a more human capital is everybody’s business. If a child prosperous and inclusive future. 116 References References Acemoglu, D., Chernozhukov, V., Werning, I., and RISE Working Paper Series. 20/039. https://doi. Whinston, M. D. (2020). “Optimal Targeted org/10.35489/BSG-RISE-WP_2020/039. Lockdowns in a Multi-Group SIR Model,” NBER Angrist, N., Djankov, S., Goldberg, P., Patrinos, H. Working Paper 27102, National Bureau of Eco- A. (2019). “Measuring Human Capital,” SSRN nomic Research, Cambridge, MA. Scholarly Paper, ID 3339416 Acosta, P. and Velarde, R. (2015). “Philippine Social Asanov, I., Flores, F., Mckenzie, D. J., Mensmann, Protection Note,” October, No. 8. World Bank, M., and Schulte, M. (2020). “Remote-learning, Washington D.C. Time-Use, and Mental Health of Ecuadorian Alatas, V., Banerjee, A., Hanna, R., Olken, B. A., Pur- High-School Students during the COVID-19 namasari, R., and Wai-Poi, M. (2019). “Does Quarantine (No. WPS9252; pp. 1–25),” World Elite Capture Matter? Local Elites and Tar- Bank, Washington D.C. geted Welfare Programs in Indonesia,” AEA Audit Office of Morocco (2017). “Progress Report Papers and Proceedings, vol 109, pages 334-339. for FY 2016 and FY 2017, Volume I.” In: Ben- Alderman, H., Hoddinott, J., and Kinsey, B. (2006). kassmi, M. and Abdelkhalek, T. (2020), Building “Long term consequences of early childhood Human Capital - Lessons from Country Experiences: malnutrition,” Oxford Economic Papers, 58(3), Morocco, pg 15-16, World Bank, Washington D.C. 450–474. Azevedo, J. P., Sanfelice, V., and Nguyen, M. (2012). Almond, D. (2006). “Is the 1918 influenza pan- “Shapley Decomposition by Components of demic over? Long-term effects of in utero a Welfare Aggregate.” Munich Personal RePEc influenza exposure in the post-1940 US pop- Archive. MPRA Paper No. 85584. https:// ulation,” Journal of Political Economy, 114(4), mpra.ub.uni-muenchen.de/85584/ 672–712. Azevedo, J. P., Inchaust, G., and Sanfelice, V. (2013). Almond, D., and Currie, J. (2011). “Killing Me “Decomposing the Recent Inequality Decline Softly: The Fetal Origins Hypothesis,” Journal in Latin America,” World Bank Policy Research of Economic Perspectives, 25(3), 153–172. Working Paper, No. 6715. Alon, T., Doepke, M., Olmstead-Rumsey, J., and Azevedo, J. P., Hazan, A., Goldemberg, D., Iqbal, Tertilt, M. (2020). “The Impact of COVID- S. A., and Geven, K. (2020). “Simulating the 19 on Gender Equality.” NBER Working potential impacts of COVID-19 school clo- Paper 26947, National Bureau of Economic sures on schooling and learning outcomes: Research, Cambridge, MA A set of global estimates,” World Bank Policy Andrabi, T., Daniels, B., and Das, J. (2020). “Human Research Working Paper, No. 9284. Capital Accumulation and Disasters: Evi- Baird, S., Friedman, J., and Schady, N. (2011). “Aggre- dence from the Pakistan Earthquake of 2005,” gate Income Shocks and Infant Mortality in T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 117 the Developing World,” The Review of Econom- Center for Global Development Policy Paper ics and Statistics, 93(3), 847–856. No. 096. Washington, D.C. Bandiera, O., Buehren, N., Goldstein, M. P., Rasul, Black, S. E., Devereux, P. J., and Salvanes, K. G. I., and Smurra, A. (2019). “The Economic Lives (2007). “From the Cradle to the Labor Mar- of Young Women in the Time of Ebola: Les- ket? The Effect of Birth Weight on Adult Out- sons from an Empowerment Program,” World comes,” The Quarterly Journal of Economics, Bank Policy Research Working Paper, No. 8760. 122(1), 409–439. Bardhan, P., and Udry, C. (1999). Development Bongaarts, J. and Guilmoto, C. (2015). “How Many Microeconomics. Oxford, UK: Oxford Univer- More Missing Women? Excess Female Mor- sity Press. tality and Prenatal Sex Selection, 1970–2050,” Bassett, M. R., and Arnhold, N. (2020). “COVID- Population and Development Review. vol. 41(2), 19’s immense impact on equity in tertiary pages 241-269, June. education,” World Bank, Washington D.C. Bora, J.K. and Saikia, N. (2018). “Neonatal and Battu, H. and Bender, K. (2020). “Educational Mis- under-five mortality rate in Indian districts match in developing countries: A review of with reference to Sustainable Development the existing evidence.” In: S. Bradley and C. Goal 3: An analysis of the National Family Green (eds), The Economics of Education: A com- Health Survey of India (NFHS), 2015–2016,” prehensive overview (2nd Ed). Academic Press. PLoS ONE 13(7): e0201125. Beegle, K., Dehejia, R. H., and Gatti, R. (2006). Bossavie, L. and Kanninen, O. (2018). “What “Child labor and agricultural shocks,” Journal Explains the Gender Gap Reversal in Educa- of Development Economics, 81(1), 80–96. tional Attainment?” World Bank Policy Research Working Paper, No. 8303. Beirne, K., Doorley, K., Regan, M., Roantree, B., and Tuda, D. (2020). “The potential costs Bundervoet, T. and Fransen, S. (2018). “The edu- and distributional effect of Covid-19 related cational impact of shocks in utero: Evidence unemployment in Ireland,” Budget Perspec- from Rwanda,” Economics & Human Biology, tives. https://doi.org/10.26504/bp202101. Elsevier, vol. 29(C), pages 88-101. Bell, B., Bindler, A., and Machin, S. (2017). “Crime Case, A. and Deaton, A. (2020). “Deaths of Despair Scars: Recessions and the Making of Career and the Future of Capitalism,” Princeton, NJ: Criminals,” The Review of Economics and Statis- Princeton University Press. tics, 100(3), 392–404. Caselli, F. (2005). “Accounting for Cross-Coun- Berkman, D. S. (2002). “Malnutrition and para- try Income Differences,” In: Philippe Aghion sitic infections in infants affect cognition in & Steven Durlauf (ed.), Handbook of Economic late childhood,” Biomedicine and Pharmacother- Growth, 1st ed., vol. 1, (no chapter) pages 679- apy, 56(5). 741, Elsevier. Bhutta, Z., Akseer, N., Keats, E., Vaivada, T., Baker, Centers for Disease Control and Prevention (2020). S., Horton, S., Katz, J., Menon, P., Piwoz, E., “Coronavirus Disease 2019 (COVID-19).” Shekar, M., Victora, C., and Black, R. (2020) Centers for Disease Control and Prevention (2019). “How countries can reduce child stunting at “CDC in eSwatini Factsheet.” scale: lessons from exemplar countries,” The Chang, H.-J., Huang, N., Lee, C.-H., Hsu, Y.-J., American Journal of Clinical Nutrition. Hsieh, C.-J., and Chou, Y.-J. (2004). “The Birdsall, N., Bruns, B., and Madan, J. (2016). “Learn- impact of the SARS epidemic on the utiliza- ing Data for Better Policy: A Global Agenda,” tion of medical services: SARS and the fear of 118 References SARS,” American Journal of Public Health, 94(4), Bank Human Capital Index,” World Bank Policy 562–564. Research Working Paper, No. 9020. Chioda L. (2016). “Work and Family: Latin Amer- Dammert, A. C., Galdo, J. C., and Galdo, V. ican and Caribbean Women in Search of a (2014). “Preventing Dengue Through Mobile New Balance,” Latin American Development Phones: Evidence from a Field Experiment Forum. World Bank, Washington D.C in Peru,” Journal of Health Economics, 35, 147- Cooper, H., Nye, B., Charlton, K., Lindsay, J., and 161. Greathouse, S. (1996). “The Effects of Sum- Datar, A., Kilburn, R., and Loughran, D. S. (2010). mer Vacation on Achievement Test Scores: A “Endowments and Parental Investments in Narrative and Meta-Analytic Review,” Review Infancy and Early Childhood,” Demography, of Educational Research, 66(3), 227–268. 47(1), 145–162. Corral, P. and Gatti, R. (2020). “Accumulation Davis, S. J., and von Wachter, T. M. (2011). “Reces- interrupted: COVID-19 and human capital sions and the Cost of Job Loss,” .” NBER Work- among the young.” COVID-19 in Developing ing Paper 17638, National Bureau of Eco- Economies, VOX EU/CEPR. https://voxeu.org/ nomic Research, Cambridge, MA content/covid-19-developing-economies De la Fuente, A., Foster, E. and Jacoby, H. (2019). Corral, P., Irwin, A., Krishnan, N., Mahler, D. G., “The Immediate Effects of the Free Quality and Vishwanath, T. (2020). “Fragility and Education Program in Sierra Leone,” World Conflict: On the Front Lines of the Fight Bank, Washington, D.C. against Poverty,” February 27, World Bank, De la Fuente, A., Jacoby, H. G., and Lawin, K. G. Washington D.C. (2019). “Impact of the West African Ebola Correia, S., Luck, S., and Verner, E. (2020). “Pan- Epidemic on Agricultural Production and demics Depress the Economy. Public Health Rural Welfare: Evidence from Liberia,” World Interventions Do Not: Evidence from the 1918 Bank, Washington D.C. Flu,” April 2, SSRN Electronic Journal. de Onis, M., and Branca, F. (2016). “Childhood COVID-19 Task Force del Dipartimento Malat- stunting: a global perspective,” Maternal & tie Infettive e Servizio di Informatica, Isti- child nutrition, 12 Suppl 1(Suppl 1), 12–26. tuto Superiore di Sanità (2020). “Epidemia Dean, T., Shane M. Murphy, and Martin E. Sand- COVID-19, Aggiornamento nazionale: 23 buc (2016). “Why Has Under-5 Mortality aprile 2020.” Decreased at Such Different Rates in Differ- Crimmins, E., Shim, H., Zhang, Y.S., Kim, J. (2019). ent Countries?” Journal of Health Economics, 48 “Differences between Men and Women in Mor- ( July): 16–25. tality and the Health Dimensions of the Mor- Department of Statistics Singapore (2020), Sing- bidity Process,” Clinical Chemistry, 65(1):135- Stat Table Builder. 145. doi:10.1373/clinchem.2018.288332 Dingel, J. I., and Neiman, B. (2020). “How many Cruz, R., Loureiro, L. (2020). “Achieving World- jobs can be done at home?” NBER Working Class Education in Adverse Socioeconomic Paper 26948, National Bureau of Economic Conditions: Research, Cambridge, MA. The Case of Sobral in Brazil,” World Bank, Wash- Drake, L., Woolnough, A., Burbano, C. and Bundy, ington D.C. D. (2016). “Global school feeding sourcebook: D’Souza, R., Gatti, R., and Kraay, A. (2019). “A lessons from 14 countries (English),” World Socioeconomic Disaggregation of the World Bank Group, Washington D.C. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 119 Dunbar, A., Hossler, D., Shapiro, D., Chen, J., Mar- Filmer, D., Rogers, H., Angrist, N., Sabarwal, S. tin, S., Torres, V., Zerquera, D., and Ziskin, M. (2018) “Learning Adjusted Years of School- (2011). “National Postsecondary Enrollment ing (LAYS): Defining a New Macro Measure Trends: Before, during, and after the Great of Education,” World Bank Policy Research Recession,” Signature Report, National Stu- Working Paper, No 8591. dent Clearinghouse. Finch, B. K., Thomas, K., and Beck, A. N. (2019). Ekhator-Mobayode, U.E. and Abebe Asfaw, A. “The Great Recession and adverse birth out- (2019). “The child health effects of terrorism: comes: Evidence from California, USA,” SSM evidence from the Boko Haram Insurgency - Population Health, 9, 100470. in Nigeria,” Applied Economics, vol. 51, issue 6, Foster, A. and Rosenzweig, M. (2010). “The Micro- 624-638. economics of Technology Adoption,” Annual El-Kogali, S. and Krafft, C. (2020). “Expectations Review of Economics, Vol. 2: 395-424 and Aspirations: A New Framework for Edu- Galal, A., Welmond, M., Carnoy, M., Nellemann, cation in the Middle East and North Africa,” S., Keller, J.; Wahba, J. and Yamasaki, I. (2008). World Bank, Washington D.C. “The road not traveled: education reform in Elston, J. W. T., Cartwright, C., Ndumbi, P., and the Middle East and North Africa (English),” Wright, J. (2017). “The health impact of the MENA development report, World Bank, 2014–15 Ebola outbreak,” Public Health, 143, Washington D.C. 60–70. Galasso, E., and Wagstaff, A. (2019). “The aggregate European Commission and World Bank (2018). income losses from childhood stunting and “A Proposed Early Warning Mechanism for the returns to a nutrition intervention aimed Early School Leaving Prevention and Pro- at reducing stunting,” Economics and Human posed Action Plan for Implementation,” Biology, 34, 225–238. Administration Agreement SRSS/S2018/063, Galea, S., Merchant, R. M., and Lurie, N. (2020). European Commission, Brussels; World “The Mental Health Consequences of COVID- Bank, Washington, D.C. 19 and Physical Distancing: The Need for Pre- Fernandez, L., and Olfindo, R. (2011). “Overview vention and Early Intervention,” JAMA Inter- of the Philippines’ Conditional Cash Transfer nal Medicine, 180(6), 817–818. Program: The Pantawid Pamilyang Pilipino Gatti, R., Morgandi, M., Grun, R., Brodmann, S., Program.” Angel-Urdinola, D., Moreno, J. M., Marotta, Figari, F., and Fiorio, C. V. (2020). “Welfare resil- D., Schiffbauer, M., and Mata Lorenzo, E. ience in the immediate aftermath of the (2013). “Jobs for Shared Prosperity: Time for covid-19 outbreak in Italy,” EUROMOD at the Action in the Middle East and North Africa,” Institute for Social and Economic Research. World Bank, Washington D.C. Filmer, D. and Pritchett, L. H. (1998). “Estimating Gelder, N. van, Peterman, A., Potts, A., O’Donnell, Wealth Effects without Expenditure Data— M., Thompson, K., Shah, N., and Oertelt-Pri- or Tears: With an Application to Educational gione, S. (2020). “COVID-19: Reducing the Enrollments in States of India,” World Bank risk of infection might increase the risk of Policy Research Working Paper, No. 1994. intimate partner violence,” EClinicalMedicine, Filmer, D. and Fox, L. (2014). “Youth Employment 21, 100348. in Sub-Saharan Africa,” World Bank, Wash- Gelli, A., Aurino, E., Folson, G., Arhinful, D., ington D.C. Adamba, C., Osei-Akoto, I., Masset, E., Watkins, 120 References K., Fernandes, M., Drake, L., and Alderman, and Economic History,” Investment in Wom- H. (2019). “A School Meals Program Imple- en’s Human Capital and Economic Develop- mented at Scale in Ghana Increases Height- ment, edited by T.P. Schultz, pg. 61-90, Uni- for-Age during Midchildhood in Girls and in versity of Chicago Press. Children from Poor Households: A Cluster Grantham-McGregor, S., Cheung, Y. B., Cueto, S., Randomized Trial,” The Journal of Nutrition, Glewwe, P., Richter, L., and Strupp, B. (2007). Volume 149, Issue 8, Pages 1434–1442. “Developmental potential in the first 5 years Gender Data Portal, database, World Bank, Wash- for children in developing countries,” The ington D.C., https://datatopics.worldbank.org/ Lancet, 369(9555), 60–70. gender/. Gribble, J. and Voss, M.L. (2009). “Family Planning Gibbs, L., Nursey, J., Cook, J., Ireton, G., Alke- and Economic Well-Being: New Evidence made, N., Roberts, M., Gallagher, H. C., Bry- from Bangladesh,” Policy Brief (May), Popu- ant, R., Block, K., Molyneaux, R., and Forbes, lation Reference Bureau, Washington, DC. D. (2019). “Delayed Disaster Impacts on Aca- GSMA (2020), “The Mobile Gender Gap Report demic Performance of Primary School Chil- 2020,” GSMA Intelligence. dren,” Child Development, 90(4), 1402–1412. GSMA. (2020). The Mobile Gender Gap Report Giles, J. and F. Witoelar (2016). Personal communi- 2020. GSMA Connected Women. https:// cation with Galasso and Wagstaff (2019). www.gsma.com/mobilefordevelopment/ Glewwe, P., Jacoby, H. G., and King, E. M. (2001). wp-content/uploads/2020/05/GSMA-The- “Early childhood nutrition and academic Mobile-Gender-Gap-Report-2020.pdf achievement: A longitudinal analysis,” Journal Guerra, R.O., Alvarado, B.E. and Zunzunegui, of Public Economics, 81(3), 345–368. M.V., (2008). “Life course, gender and ethnic Glewwe, Paul, Jongwook Lee, Khoa Vu, and Hai Anh inequalities in functional disability in a Bra- Dang (2017). “What explains Vietnam’s excep- zilian urban elderly population,” Aging clinical tional performance in education relative to and experimental research, 20(1), pp.53-61. other countries? Analysis of the 2012 PISA data. Hale, T., Webster, S., Petherick, A., Phillips, T., and ” RISE annual conference, Center for Global Kira, B. (2020). “Oxford COVID-19 Govern- Development, Washington, DC, pp. 15-16. ment Response Tracker,” Blavatnik School of Global Burden of Disease (GBD), database, Institute Government. for Health Metrics and Evaluation (IHME), Handel, M., Valerio, A. and Sanchez Puerta, M. Seattle, http://www.healthdata.org/gbd. (2016). “Accounting for mismatch in low- and Global Jobs Indicators Database ( JOIN) database. middle-income countries: measurement, https://datacatalog.worldbank.org/dataset/glob- magnitudes, and explanations,” Directions in al-jobs-indicators-database development, World Bank, Washington D.C. Global System for Mobile Communications Hanushek, Eric A., and Ludger Woessmann. 2011. (GSMC) (2020). “Sample selectivity and the validity of interna- Golberstein, E., Wen, H., and Miller, B. F. (2020). tional student achievement tests in economic “Coronavirus Disease 2019 (COVID-19) and research,” Economics Letters 110(2): 79-82. Mental Health for Children and Adolescents.” Health Equity and Financial Protection Indica- JAMA Pediatrics. tors (HEPHI), database, World Bank, Wash- Goldin, C. (1995).“The U-Shaped Female Labor ington D.C. http://datatopics.worldbank.org/ Force Function in Economic Development health-equity-and-financial-protection/ T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 121 Hershbein, B. J. (2012). “Graduating High School Jamison, D. T., Shane M. Murphy, and Martin E. in a Recession: Work, Education, and Home Sandbuc (2016). “Why Has Under-5 Mortality Production,” The B.E. Journal of Economic Anal- Decreased at Such Different Rates in Differ- ysis & Policy, 12(1). ent Countries?” Journal of Health Economics, 48 ( July): 16–25. Hobbs, J. E. (2020). “Food supply chains during the COVID-19 pandemic,” Canadian Jour- Joint Child Malnutrition Estimates ( JME): Levels nal of Agricultural Economics/Revue Canadienne and trends (2020 edition), database, https:// d’agroeconomie. data.unicef.org/resources/jme-report-2020. Hogan, A., Jewell, B., Sherrard-Smith, E., Vesga, J., Junquera-Varela, R.F., Verhoeven, M., Shukla, G. P., Watson, O., Whittaker, C., Hamlet, A., Smith, Haven, B., Awasthi, R. and Moreno-Dodson, J., Ainslie, K., Baguelin, M., Bhatt, S., Boonya- B. (2017). “Strengthening Domestic Resource siri, A., Brazeau, N., Cattarino, L., Charles, G., Mobilization Moving from Theory to Prac- Cooper, L., Coupland, H., Cuomo-Dannen- tice in Low- and Middle-Income Countries.” burg, G., Dighe, A., Hallett, T. (2020). “Report World Bank, Washington D.C. 19: The potential impact of the COVID-19 Kaffenberger, M. and Pritchet, L. (2017). “More epidemic on HIV, TB and malaria in low- and school or more learning? Evidence from middle-income countries.” The Lancet Global learning profiles from the financial inclusion Health, Volume 8, Issue 9. insights data,” World Development Report Back- Hoogeveen, J. and Pape, U. (2020). “Data Collec- ground Paper. World Bank, Washington D.C. tion in Fragile States: Innovations from Africa Kim, J. (2018). “The Human Capital Gap: Getting and Beyond,” Open Knowledge Repository, Governments to Invest in People,” Foreign World Bank, Washington D.C. Affairs, July/August 2018. ILO (International Labour Organization) (2018). Klasen, S. (2019). “What Explains Uneven Female “Global Wage Report 2018/19: What Lies Labor Force Participation Levels and Trends Behind Gender Pay Gaps,” International in Developing Countries?” World Bank Labour Office. Research Observer, 34(2). ILO (2020a). “As job losses escalate, nearly half of Kolen, M. J., & Brennan, R. L. (2004). “Test Equat- global workforce at risk of losing livelihoods,” ing, Scaling, and Linking: Methods and Prac- Press release. tices (2nd ed.),” Springer-Verlag. https://doi. ILO (2020b). “Impact of lockdown measures on org/10.1007/978-1-4757-4310-4 the informal economy,” Press release. Kraay, A. (2018). “Methodology for a World Bank ILOSTAT. Retrieved from World Bank Gender Human Capital Index,” World Bank Policy Data Portal. Research Working Paper, No. 8593. Institut international de planification de l’édu- Krafft C., Sieverding M., Keo C. And Salemi C. (2018). cation, United Nations Educational, Scien- “Syrian Refugees in Jordan: Demographics, tific, and Cultural Organization and UNICEF Livelihoods, Education, and Health,” Economic (2018). “République du Mali: Analyse du sec- Research Forum Working Paper No. 1184. teur de l’éducation,” Dakar, Senegal, IIPE– Laborde, D. (2020). IFPRI: Food Export Restric- Pôle de Dakar. tions Tracker-Tableau Dashboards. Tableau International Monetary Fund (2020). “Global Software. Prospects and Policies,” In World Economic LaFave, D., and Thomas, D. (2017). “Height and Outlook, April 2020: The Great Lockdown. cognition at work: Labor market productivity 122 References in a low-income setting.” Economics & Human Intervention Improved Adult Human Capi- Biology, 25, 52–64. tal and Economic Productivity,” The Journal of Nutrition, 140(2), 411–414. Lahti-Pulkkinen, M., Bhattacharya, S., Räikkönen, K., Osmond, C., Norman, J. E., and Reynolds, Mary, S. (2018). “How much does economic growth R. M. (2018). “Intergenerational Transmission contribute to child stunting reductions?” of Birth Weight across Three Generations,” Economies, 6(4), 55. American Journal of Epidemiology, 187(6), 1165– Mary, S., Shaw, K., & Paloma, S. G. y. (2019). “Does 1173. the sectoral composition of growth affect Lain, J., Perng, J., Vishwanath, T., Azad, MA., Siwatu, child stunting reductions?” Development Policy GO., Palacio-Lopez, A., McGee, K., Amankwa, Review, 37(2), 225–244. A., Contreras, I., (2020). “COVID-19 From McDonnell, A., Urrutia, A., Samman, E. (2019). the Ground Up: What the Crisis Means for “Reaching universal health coverage: a politi- Nigerians: Monitoring COVID-19 Impacts on cal economy review of trends across 49 coun- Households in Nigeria,” World Bank, Wash- tries,” ODI Working Paper No. 570, London, ington D.C. UK. Lancker, W. V., and Parolin, Z. (2020). “COVID-19, Merotto, D., Weber, M. and Reyes, A. (2018). “Path- school closures, and child poverty: A social ways to Better Jobs in IDA Countries: Find- crisis in the making,” The Lancet Public Health, ings from Jobs Diagnostics,” Jobs Series; No. 5(5), e243–e244. 14. World Bank, Washington D.C. Le Nestour, A., and Moscoviz, L. (2020). “Five Meyers, K., and Thomasson, M. (2017). “Paralyzed Findings from a New Phone Survey in Sene- by Panic: Measuring the Effect of School Clo- gal.” Center For Global Development. sures during the 1916 Polio Pandemic on Ma, L., Shapira, G., de Walque, D., Do, Q.-T., Fried- Educational Attainment,” NBER Working man, J., and Levchenko, A. A. (2020). “The Paper 23890, National Bureau of Economic intergenerational dilemma of COVID-19 lock- Research, Cambridge, MA. down policies,” World Bank, Washington D.C. Miller, G., and Urdinola, B. P. (2010). “Cyclicality, Mahler, D. G., Lakner, C., Castaneda, A., and Wu, Mortality, and the Value of Time: The Case of H. (2020a). “The impact of COVID-19 (Coro- Coffee Price Fluctuations and Child Survival navirus) on global poverty: Why Sub-Saharan in Colombia,” The Journal of Political Economy, Africa might be the region hardest hit,” World 118(1), 113–155. Bank blogs, Washington D.C. Ministry of Education and Science, Republic of Mahler, D. G., Lakner, C., Castaneda, A., and Wu, Albania (2005). “National Education Strategy H. (2020b). “Updated estimates of the impact 2004-2015.” of COVID-19 on global poverty,” World Bank Minoiu, C. and Shemyakina, O. N. (2014). “Armed blogs, Washington D.C. conflict, household victimization, and child Mahmud, M., & Riley, E. (forthcoming). “House- health in Côte d’Ivoire,” Journal of Development hold Response to an Extreme Shock: Evi- Economics, 108, (C), 237-255. dence on the Immediate Impact of the Nelson, R. (2020). “COVID-19 disrupts vaccine Covid-19 Lockdown on Economic Outcomes delivery.” The Lancet Infectious Diseases, 20(5), and Well-being in Rural Uganda”. 546. Martorell, R., Melgar, P., Maluccio, J. A., Stein, A. Newbrander, W., Waldman, R. and Shepherd-Ban- D., and Rivera, J. A. (2010). “The Nutrition igan, M. (2011). “Rebuilding and strengthening T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 123 health systems and providing basic health ser- Philip, R. K., Purtill, H., Reidy, E., Daly, M., Imcha, vices in fragile states,” Disasters, 35: 639-660. M., McGrath, D., O’Connell, N. H., & Dunne, C. P. (2020). “Reduction in preterm births O’Hare, B., Makuta, I., Chiwaula, L., and Bar- during the COVID-19 lockdown in Ireland: Zeev, N. (2013). “Income and child mortality A natural experiment allowing analysis of in developing countries: A systematic review data from the prior two decades.” MedRxiv, and meta-analysis,” Journal of the Royal Society 2020.06.03.20121442 of Medicine, 106(10), 408–414. Poletaev, M., and Robinson, C. (2008). “Human Olivetti, C. and Petrongolo, B. (2017). “The Eco- Capital Specificity: Evidence from the Dic- nomic Consequences of Family Policies: tionary of Occupational Titles and Displaced Lessons from a Century of Legislation in Worker Surveys, 1984-2000,” Journal of Labor High-Income Countries,” Journal of Economic Economics, 26(3), 387–420. Perspectives, 31 (1): 205-30. Qualls, N., Levitt, A., Kanade, N., Wright-Jegede, Olsen, A. S., and Prado, J. (2020). “COVID-19 y la N., Dopson, S., Biggerstaff, M., Reed, C., Uzic- transición de la educación privada a la pública anin, A., Levitt, A., Dopson, S., Frank, M., Hol- en Ecuador,” Enfoque Educación. loway, R., Koonin, L., Rasmussen, S., Redd, S., Oreopoulos, P., von Wachter, T., and Heisz, A. (2012). de la Motte Hurst, C., Kanade, N., Qualls, N., “The Short-and Long-Term Career Effects of Rainey, J., Reed, C. (2017). “Community Miti- Graduating in a Recession,” American Economic gation Guidelines to Prevent Pandemic Influ- Journal: Applied Economics, 4(1), 1–29. JSTOR. enza—United States, 2017.” MMWR Recom- Packard, T., Gentilini, U., Grosh, M., O’Keefe, P., mendations and Reports, 66(1), 1–32. Palacios, R., Robalino, D., and Santos, I. (2019). Quisumbing, A., Meinzen-Dick, R., Behrman, J., “Protecting all: Risk sharing for a diverse and and Basset, L. (2011). “Gender and the global diversifying world of work.” World Bank, food-price crisis,” Development in Practice, Washington D.C. 21(4–5), 488–492. Patrinos, H. A. and Angrist, N. (2018) “Global Ravallion, M. (2010). “Poverty Lines Across the Dataset on Education Quality: A Review World,” World Bank Policy Research Working and Update (2000 - 2017),” World Bank Policy Paper, No. 5284. Research Working Paper, No. 8592. Ravindran, S., and Shah, M. (2020). “Unintended Pennings, S. (2020). “The Utilization-adjusted Consequences of Lockdowns: COVID-19 Human Capital Index (UHCI)”. World Bank and the Shadow Pandemic,” NBER Working Policy Research Working Paper, No. 9375. Paper 17638, National Bureau of Economic Perez-Vincent, S., Carreras, E., Gibbons, M. A., Research, Cambridge, MA. Murphy, T. E., and Rossi, M. (2020). “COVID- Reimers, F., and A. Schleicher (2020). “Schooling 19 Lockdowns and Domestic Violence: Evi- Disrupted. How the COVID-19 pandemic is dence from Two Studies in Argentina,” Changing Education,” OECD. Working Papers, No. 143. Universidad de San Roberton, T., Carter, E. D., Chou, V. B., Stegmuller, Andres, Departamento de Economia. A. R., Jackson, B. D., Tam, Y., Sawadogo-Lewis, Phadera, L. (2019). “Unfortunate Moms and T., and Walker, N. (2020). “Early estimates of Unfortunate Children: Impact of the Nepali the indirect effects of the COVID-19 pan- Civil War on Women’s Stature and Intergen- demic on maternal and child mortality in erational Health (English),” World Bank Policy low-income and middle-income countries: A Research Working Paper, No. 8927. modelling study.” The Lancet Global Health. 124 References Rosales-Rueda, M. (2018). “The impact of early of Syrian Refugees in Jordan,” Economic life shocks on human capital formation: Evi- Research Forum Working Paper No. 1261. dence from El Niño floods in Ecuador.” Jour- Smith, O. and Nguyen, S. (2013). “Getting better: nal of Health Economics, 62, 13–44. Improving Health System Outcomes in Europe Rose, E. (2000). “Gender bias, credit constraints and Central Asia,” World Bank, Washington D.C. and time allocation in rural India,” The Eco- Stiglitz, J., Sen, A. and Fitoussi, J.P. (2009). “The nomic Journal, 110(465), 738–758. measurement of economic performance Rothstein, J. (2020). “The Lost Generation? Labor and social progress revisited: Reflections and Market Outcomes for Post Great Recession Overview.” Sciences Po, Publications 2009-33. Entrants,” ,” NBER Working Paper 27516, Sullivan, D., and von Wachter, T. (2009). “Job dis- National Bureau of Economic Research, placement and mortality: An analysis using Cambridge, MA. administrative data.” The Quarterly Journal of Ruel, M. T., Alderman, H., and Group, M. and C. Economics, 124(3), 1265–1306. N. S. (2013). “Nutrition-sensitive interventions Sumberg, J. and Sabates-Wheeler, R. (2011). “Link- and programmes: How can they help to accel- ing agricultural development to school feed- erate progress in improving maternal and child ing in sub-Saharan Africa: Theoretical per- nutrition?” The Lancet, 382(9891), 536–551. spectives,” Food Policy, 36, (3), 341-349. Saha, T., and Bhattacharya, S. (2020). “Consequence Task force COVID-19 del Dipartimento Malat- of Lockdown amid Covid-19 Pandemic on tie Infettive e Servizio di Informatica, Istituto Indian Agriculture,” Food and Scientific Reports. Superiore di Sanità. (2020). “Epidemia COVID- Savasi, V. M., Parisi, F., Patanè, L., Ferrazzi, E., 19,” Aggiornamento nazionale, April 2020. Frigerio, L., Pellegrino, A., Spinillo, A., Tateo, Tesfaye, A., Habte, Y., and Minten, B. (2020). S., Ottoboni, M., Veronese, P., Petraglia, F., “COVID-19 is shifting consumption and dis- Vergani, P., Facchinetti, F., Spazzini, D., and rupting dairy value chains in Ethiopia,” Inter- Cetin, I. (2020). “Clinical Findings and Disease national Food Policy Research Institute (IFPRI) Severity in Hospitalized Pregnant Women Tiltnes, A., Zhang, H. and Pedersen, J. (2019), “The With Coronavirus Disease 2019 (COVID-19),” living conditions of Syrian refugees in Jordan. Obstetrics & Gynecology, 136(2), 252–258. Results from the 2017-2018 survey of Syrian Schady, N. R. (2004). “Do Macroeconomic Crises refugees inside and outside camps.” Relief Web, Always Slow Human Capital Accumulation?” Fafo Institute Government of Jordan. The World Bank Economic Review, 18(2), 131– Ubdaidur, R., Phillips, J.F., Jyotsnarnoy 154. JSTOR. Chakraborty, and Koenig, M. A. (1987). “The Schmidhuber, J. (2020). “COVID-19: From a Use Effectiveness of the Copper T-200 in Global Health Crisis to a Global Food Crisis?” Matlab.” International Journal of Gynecology and Food Outlook, June 2020. Obstetrics, 25 (4): 315–22. Shafiq, M. N. (2010). “The Effect of an Economic UN Data, database, https://data.un.org/ Crisis on Educational Outcomes: An Eco- UN High Commissioner for Refugees (UNHCR, nomic Framework and Review of the Evi- 2017). “Left Behind: Refugee Education in dence,” SSRN Scholarly Paper, ID 1709045 Crisis”. Sieverding, M., Krafft, C., Berri, N., Keo, C. and UNAIDS (2020). Eswatini Country Profile. Sharpless, M. (2018). “Education Interrupted: Enrollment, Attainment, and Dropout UNAIDS (2020a). Zimbabwe Country Profile. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 125 UNESCO (2018), “TCG4 Measurement Strategies,” 2018 Health Equity and Financial Protection UNESCO Institute for Statistics. Indicators Database,” Lancet Global Health, 7 UNESCO (2019). “Global Education Monitoring [1]: E22–E23). Report 2019: Migration, Displacement and Edu- Walker, S. P., Chang, S. M., Powell, C. A., and cation – Building Bridges, not Walls,” UNESCO. Grantham-McGregor, S. M. (2005). “Effects of UNESCO (2020). “Covid-19: Education: From dis- early childhood psychosocial stimulation and ruption to recovery,” UNESCO. nutritional supplementation on cognition and education in growth-stunted Jamaican children: UNFPA. (2020). “Impact of the COVID-19 Pan- Prospective cohort study,” The Lancet, 366(9499), demic on Family Planning and Ending Gen- 1804–1807. der-based Violence, Female Genital Mutila- tion and Child Marriage.” Wenham, C., Smith, J., and Morgan, R. (2020). “COVID-19: The gendered impacts of the UNICEF (2013). “A Passport to Protection: A Guide outbreak,” The Lancet, 395(10227), 846–848. to Birth Registration Programming,” UNICEF, New York. World Bank (2012). “What Matters Most for School Health and School Feeding: A Framework UNICEF (2019). “Education under Threat in Paper,” World Bank, Washington D.C. West and Central Africa.” UNICEF Child Alert (August), UNICEF, New York. World Bank (2013). “Jobs for Shared Prosperity: Time for Action in the Middle East and North UNICEF (2020). “Eswatini HIV/AIDS Programme: Africa,” World Bank, Washington D.C. HIV prevention and response for children and adolescents,” UNICEF, New York. World Bank (2016). “Rewarding Provider Perfor- mance to Improve Quality and Coverage of United Nations (2011). Department of Economic Maternal and Child Health Outcomes: Zim- and Social Affairs, Population Division. babwe Results-Based Financing Pilot Pro- United Nations Interagency Group for Child Mor- gram,” World Bank, Washington D.C. tality Estimation (UNIGME, 2019). “Levels World Bank (2016a). “World Development Report and Trends in Child Mortality, Report 2019,” 2016: Digital Dividends,” World Bank, Wash- United Nations Children’s Fund, New York. ington D.C. United Nations Statistics Division (2014). “Coverage of World Bank (2018). “World Development Report Birth and Death Registration,” United Nations, 2018: Learning to Realize Education’s Prom- Demographic and Social Statistics, Civil Regis- ise,” World Bank, Washington D.C. tration and Vital Statistics. https://unstats.un.org/ unsd/demographic-social/crvs/#coverage World Bank (2018a). “The Human Capital Project,” World Bank, Washington D.C. Vogl, T. S. (2014). “Height, skills, and labor market outcomes in Mexico,” Journal of Development World Bank (2018b). “Progress in the Face of Economics, 107, 84–96. Insecurity: Improving Health Outcomes in Afghanistan,” World Bank, Washington D.C. Von Hippel, P. T. , and Hamrock, C. (2019). “Do test score gaps grow before, during, or between World Bank (2018c). “Reducing Childhood Stunt- the school years? Measurement artifacts and ing with a New Adaptive Approach,” World what we can know in spite of them,” Sociologi- Bank, Washington D.C. cal Science, 6, 43–80. World Bank (2018d) “Human Capital: A Project for Wagstaff, A., Eozenou, P., Neelsen, S. and Smitz, the World,” Development Committee, World M-F (2019). “Introducing the World Bank’s Bank, Washington, D.C. 126 References World Bank (2019). “World Development Report: World Development Indicators (WDI), data- The changing nature of work,” World Bank, base, https://datacatalog.worldbank.org/dataset/ Washington D.C. world-development-indicators World Bank (2019a). “Ending Learning Poverty: What World Food Programme (2013). “State of School Will It Take?” World Bank, Washington D.C. Feeding Worldwide,” World Food Programme (WFP), New York. World Bank (2019b). “Insights from Disaggregat- ing the Human Capital Index,” World Bank, World Food Programme (2017). “Fill the Nutri- Washington D.C. ent Gap: Pakistan. Summary Report,” World Food Programme (WFP), New York. World Bank (2019c). “Rwanda Nutrition Situa- tion Analysis and Policy Implications,” World World Food Programme (2019). “Cost-effective & Bank, Washington D.C. sector-specific recommendations for improv- ing nutrition outcomes in Pakistan: Evidence World Bank (2019d). “Improving Access to Maternal from the Fill the Nutrient Gap Analysis,” World Health for Zimbabwe’s Expectant Mothers,” Food Programme (WFP), New York, USA. Feature Story, World Bank, Washington D.C. World Health Organization (2009). “The WHO World Bank (2020). “Global Economic Prospects”, Multicentre Growth Reference Study World Bank, Washington D.C. (MGRS),” World Health Organization (WHO), World Bank (2020a). “The Effect of the H1N1 Geneva, Switzerland. pandemic on learning. What to expect with World Health Organization (2020a). “At least 80 Covid-19?” World Bank, Washington D.C. million children under one at risk of dis- World Bank (2020b). “Protecting the Poor and Vulner- eases such as diphtheria, measles and polio able: Social Response Framework for COVID- as COVID-19 disrupts routine vaccination 19,” Social Protection and Jobs Global Practice. efforts, warn Gavi, WHO and UNICEF,” World World Bank (2020c). “COVID-19 and Health Sys- Health Organization (WHO). tems: Just the Beginning,” World Bank, Wash- World Health Organization (2020b). “Stunting, ington D.C. wasting, overweight and underweight,” World World Bank (2020d). “Protecting People and Health Organization (WHO). Economies: Integrated Policy Responses to World Population Prospects 2019, database, COVID-19.” World Bank, Washington D.C. https://population.un.org/wpp/ World Bank (2020e). “World Bank Group Strategy Wren-Lewis, S. (2020). “The economic effects of a for Fragility, Conflict, and Violence 2020– pandemic,” In Economics in the Time of COVID- 2025 (English),” World Bank, Washington 19, 109–112. Centre for Economic Policy D.C. Research (CEPR). World Bank (2020f). “Markets and People: Roma- Yang, H., and de Onis, M. (2008). “Algorithms for nia Country Economic Memorandum,” converting estimates of child malnutrition Chapter 4. World Bank, Washington, D.C. based on the NCHS reference into estimates World Bank (forthcoming). “Reform Priorities for based on the WHO Child Growth Standards,” Pre-University Education in Albania,” World BMC Pediatr, 8 (19). Bank, Washington D.C. Yourish, K., Lai, K. K. R., Ivory, D., and Smith, World Bank Microdata Catalogue, database, M. (2020). “One-Third of All U.S. Coronavi- https://microdata.worldbank.org/index.php/ rus Deaths Are Nursing Home Residents or catalog Workers,” The New York Times, May 11, 2020. APPENDICES Appendix A: The Human Capital Index: Methodology T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 131 1. COMPONENTS OF THE HCI The Human Capital Index (HCI) measures the human capital that a child born today can expect to attain by age 18, given the risks to poor health and poor education that prevail in the country where she lives. The HCI follows the trajectory from birth to adulthood of a child born today.1 In the poorest countries in the world, there is a sig- nificant risk that the child will not survive to her fifth birthday. Even if she does reach school age, there is a further risk that she will not start school, let alone complete the full cycle of 14 years of school from preschool to grade 12 that is the norm in rich coun- tries. The time she does spend in school may translate unevenly into learning, depend- ing on the quality of the teachers and schools she experiences. When she reaches age 18, she carries with her the lasting effects of poor health and nutrition from her childhood that limit her physical and cognitive abilities as an adult. The HCI quantitatively illustrates the key stages in this trajectory and their conse- quences for the productivity of the next generation of workers, with three components: Component 1: Survival. This component of the index reflects the unfortunate reality that not all children born today will survive until the age when the process of human cap- ital accumulation through formal education begins. It is measured using the under-5 mortality rate (Figure A.1), with survival to age 5 as the complement of the under-5 mortality rate. Component 2: School. This component of the index combines information on the quan- tity and quality of education; · The quantity of education is measured as the number of years of school a child can expect to obtain by age 18 given the prevailing pattern of enrollment rates (figure A.1). The maximum possible value is 14 years, corresponding to the maximum number of years of school obtained as of her 18th birthday by a child who starts preschool at age 4. In the data, expected years of school range from around 4 to close to 14 years. · The quality of education reflects work at the World Bank to harmonize test scores from major international student achievement testing programs into a measure 1 This appendix provides a summary of the methodology for the Human Capital Index. For additional details, see Kraay (2018), on which this appendix is based. 132 A ppendix A: T he H u man Capita l Index : Metho d o lo gy of harmonized test scores (HTSs).2 HTSs are measured in units of the Trends in International Mathematics and Science Study (TIMSS) testing program and range from around 300 to around 600 across countries (figure A.1). Tests scores are used to convert expected years of school into quality-adjusted years of school. Quality-adjusted years of school are obtained by multiplying expected years of school by the ratio of test scores to 625, corresponding to the TIMSS benchmark of advanced achievement.3 For example, if expected years of school in a country is 10 and the average test score is 400, then the country has 10 × (400/625) = 6.4 quality-adjusted years of school. The distance between 10 and 6.4 represents a learning gap equivalent to 3.6 years of school. Component 3: Health. There is no single broadly accepted, directly measured, and widely available summary measure of health that can be used in the same way as years of school as a standard measure of educational attainment. Instead, two proxies for the overall health environment are used: Adult survival rates. This is measured as the share of 15-year-olds who survive until age 60. This measure of mortality serves as a proxy for the range of nonfatal health outcomes that a child born today would experience as an adult if current conditions prevail into the future. Healthy growth among children under age 5. This is measured as the fraction of children who are not stunted, that is, as 1 minus the share of children under 5 whose height- for-age is more than two standard deviations below the World Health Organization Child Growth Standards’ median. Stunting serves as an indicator for the prenatal, infant, and early childhood health environment, summarizing the risks to good health that children born today are likely to experience in their early years, with important consequences for health and well-being in adulthood. Data on these two health indicators are shown in figure A.1. Data for all the components of the HCI 2020 by country are reported in Table C2.3. Aggregation methodology The components of the HCI are combined into a single index by first converting them into contributions to productivity.4 Multiplying the contributions to productivity gives the overall HCI. The HCI summarizes how productive children born today will be as members of the future workforce, given the risks to education and health summarized in the components. The HCI is measured in units of productivity relative to a bench- mark corresponding to complete education and full health. 2 The methodology for harmonizing test scores is detailed in Altinok, Angrist, and Patrinos (2018) and Patrinos and Angrist (2018). 3 This methodology was introduced by the World Bank (2018) and is elaborated on in Angrist et al. (2019). 4 This approach has been used extensively in the development accounting literature (for example, Caselli, 2005; Hsieh and Klenow, 2010). The approach for health closely follows Weil (2007). Galasso and Wagstaff (2016) apply a similar framework to measure the costs of stunting. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 133 In the case of survival, the relative productivity interpretation is stark: children who do not survive childhood never become productive adults. As a result, expected productiv- ity as a future worker of a child born today is reduced by a factor equal to the survival rate, relative to the benchmark where all children survive. In the case of education, the relative productivity interpretation is anchored in the large empirical literature measuring the returns to education at the individual level. A rough consensus from this literature is that an additional year of school raises earnings by about 8 percent.5 This evidence can be used to convert differences in quality-adjusted years of school across countries into differences in worker productivity. For example, compared with a benchmark where all children obtain a full 14 years of school by age 18, a child who obtains only 10 years of education can expect to be 32 percent less pro- ductive as an adult (a gap of four years of education, multiplied by 8 percent per year). In the case of health, the relative productivity interpretation is based on the empirical literature measuring the economic returns to better health at the individual level. The key challenge in this literature is that there is no unique directly measured summary indicator of the various aspects of health that matter for productivity. This literature often uses proxy indicators for health, such as adult height, in the same way that the HCI uses proxy indicators.6 This is because adult height can be measured directly and reflects the accumulation of shocks to health through childhood and adolescence. A rough consensus drawn from this literature is that an improvement in health associated with a 1 centimeter increase in adult height raises productivity by 3.4 percent. Converting this evidence on the returns to one proxy for health (adult height) into the other proxies for health used in the HCI (stunting and adult survival) requires informa- tion on the relationships between these different proxies.7 For stunting, there is a direct relationship between stunting in childhood and future adult height because growth deficits in childhood persist to a large extent into adulthood, together with the associated health and cognitive deficits. Available evidence suggests that a reduction in stunting rates of 10 percentage points increases attained adult height by approximately one centimeter (0.1 × 10.2), which increases productivity by 3.5 percent. For adult survival, the empirical evidence suggests that, if overall health improves, both adult height and adult survival rates increase in such a way that adult height rises by 1.9 centimeters for every 10-percentage-point improvement in adult survival. This implies that an improvement in health that leads to an increase in adult survival rates of 10 percentage points is associated with an improvement in worker productivity of 1.9 × 3.4 percent, or 6.5 percent. 5 The seminal methodology is due to Mincer (1958). See Montenegro and Patrinos (2014) for recent cross-coun- try estimates of the returns to schooling. 6 For example, see Case and Paxson (2008); Horton and Steckel (2011). 7 For details, see Weil (2007) and Kraay (2018), section A3, and accompanying references. 134 A ppendix A: T he H u man Capita l Index : Metho d o lo gy In the HCI, the estimated contributions of health to worker productivity based on these two alternative proxies are averaged (if both are available) and are used individually (if only one of the two is available). The contribution of health to productivity is expressed relative to the benchmark of full health, defined as the absence of stunting, and a 100 percent adult survival rate. For example, compared with a benchmark of no stunting, in a country where the stunting rate is 30 percent, poor health reduces worker productivity by 30 × 0.34 per- cent, or 10 percent. Similarly, compared with the benchmark of 100 percent adult survival, poor health reduces worker productivity by 30 × 0.65 percent, or 19.5 percent, in a country where the adult survival rate is 70 percent. The average of the two estimates of the effect of health on productivity is used in the HCI. The overall HCI is constructed by multiplying the contributions of survival, school, and health to relative productivity, as follows: HCI =​_​ Survival × School × Health​, ​ (1) with the three components defined as: ≡ ___________ 1 − Under 5 ​    ​Mortality Rate ​Survival ​    ​​  ​ 1 2) ​School ≡ ​e​ϕ​(​Expected Years of School ×​   Harmonized Test Score​−14​)​ ​   (3) ___________ ​ ​ 625 Health ≡ ​e​​(​γ​ ASR​×(​ Adult Survival Rate−1)​+​γStunting ​ ​×(​ Not Stunted Rate−1)​)​/2   (4) ​ The components of the index are expressed here as contributions to productivity rela- tive to the benchmark of complete high-quality education and full health. The param- eter ​ϕ = 0.08​ measures the returns to an additional year of school. The parameters​​ γ​ASR​ = 0.65​and ​γ​Stunting​ = 0.35​measure the improvements in productivity associated with an improvement in health, using adult survival and stunting as proxies for health. The benchmark of complete high-quality education corresponds to 14 years of school and a harmonized test score of 625. The benchmark of full health corresponds to 100 percent child and adult survival and a stunting rate of 0 percent. These parameters serve as weights in the construction of the HCI. The weights are cho- sen to be the same across countries, so that cross-country differences in the HCI reflect only cross-country differences in the component variables. This facilitates the interpre- tation of the index. This is also a pragmatic choice because estimating country-specific returns to education and health for all countries included in the HCI is not feasible. As shown in figure A.1, child survival rates range from around 90 percent in the highest-mortality countries to near 100 percent in the lowest-mortality countries. This implies a loss of productivity of 10 percent relative to the benchmark of no mortality. Quality-adjusted years of school range from around 3 years to close to 14 years. This gap in quality-adjusted years of school implies a gap in productivity relative to the benchmark of complete education of ​e​ϕ​(​3−14​)​​ = ​e​0.08​(​−11​)​​ = 0.4​; that is, the productivity of a T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 135 future worker in countries with the lowest years of quality-adjusted school is only 40 percent of what it would be under the benchmark of complete education. For health, adult survival rates range from 60 to 95 percent, while the share of children not stunted ranges from around 60 percent to over 95 percent. Using adult survival rates indicates ​(​0.6−1​)​ a gap in productivity of ​e​​γ​ ASR ​ = ​e​0.65​(​−0.4​)​​ = 0.77​. Thus, based on adult survival rates as a proxy for health, the productivity of a future worker is only 77 percent of what it would be under the benchmark of full health. Using the share of children not stunted leads to Stunting ( ​ ​0.6−1​)​ a gap in productivity of ​e​​γ​ ​ = ​e​0.35​(​−0.4​)​​ = 0.87​. The productivity of a future worker using the stunting-based proxy for health is therefore only 87 percent of what it would be under the benchmark of full health. 2. THE HUMAN CAPITAL INDEX The overall HCI is displayed in figure 2.1 the main text. The HCI data are available at www.worldbank.org/humancapital. The HCI is, on average, higher in rich countries than in poor countries and ranges from around 0.3 to around 0.9. The units of the HCI have the same interpretation as the components measured in terms of relative produc- tivity. Consider a country such as Morocco, which has an HCI of around 0.5. If current education and health conditions in Morocco persist, a child born today will be only half as productive as she could have been if she enjoyed complete education and full health. All of the components of the HCI are measured with some error, and this uncer- tainty naturally has implications for the precision of the overall HCI. To capture this imprecision, the HCI estimates for each country are accompanied by upper and lower bounds that reflect the uncertainty in the measurement of the components of the HCI (figure A.2). These bounds are constructed by recalculating the HCI using lower- and upper-bound estimates of the components of the HCI. The resulting uncertainty intervals are shown in figure A.2 as vertical ranges around the value of the HCI for each country. The upper and lower bounds are a tool to highlight to users that the estimated HCI values for all countries are subject to uncertainty, reflecting the corresponding uncer- tainty in the components. In cases where these intervals overlap for two countries, this indicates that the differences in the HCI estimates for these two countries should not be overinterpreted because they are small relative to the uncertainty around the value of the index itself. This is intended to help move the discussion away from small differ- ences in country ranks on the HCI and toward more useful discussions around the level of the HCI and what this implies for the productivity of future workers. Another feature of the HCI is that it can be disaggregated for girls and boys in the 153 coun- tries where sex-disaggregated data on all of the components of the index are available. Gender gaps are most pronounced in survival to age 5, adult survival, and stunting, where girls, on average, do better than boys in nearly all countries. The number of expected years of school is higher among girls than boys in about two-thirds of the countries, as are test scores. Overall, HCI scores are higher among girls than boys in the majority of countries. 136 A ppendix A: T he H u man Capita l Index : Metho d o lo gy The HCI uses the returns to education and health to convert the education and health indicators into differences in worker productivity across countries. The higher the returns, the larger the resulting worker productivity differences. The size of the returns also influences the relative contributions of education and health to the overall index. For example, if the returns to education are high while the returns to health are low, then cross-country differences in education will account for a larger portion of cross-country differences in the index. Although varying the assumptions about the returns to education and health will affect the relative positions of countries on the index, in practice these changes are small because the health and education indicators are strongly correlated across countries.8 Connecting the Human Capital Index to future growth and income The HCI can be connected to future aggregate income levels and growth following the logic of the development accounting literature. This literature typically adopts a simple Cobb-Douglas form for the aggregate production function, as follows: ​y = A ​kp​α​​kh​1−α ​ ​​, (5) where ​y​ is GDP per worker; ​kp​ ​​ and ​kh​ ​​ are the stocks of physical and human capital per ​ ​is total factor productivity; and α worker; A ​ ​is the output elasticity of physical capital. To analyze how changes in human capital may affect income in the long run, it is useful to rewrite the production function as follows: (​ _​k​p​​)​ ​1−α​​A​​_1−α1 ​​​k​ ​ _ α ​ ​ ​y = y h (6) In this formulation, GDP per worker is proportional to the human capital stock per worker, holding constant the level of total factor productivity and the ratio of physi- cal capital to output,​​_y ​​. This formulation can be used to answer the question, “By how ​kp​ ​ much does an increase in human capital raise output per worker, in the long run after taking into account the increase in physical capital that is likely to be induced by the increase in human capital?” Equation (6) shows the answer: output per worker increases equiproportionately to human capital per worker, that is, a doubling of human capital per worker will lead to a doubling of output per worker in the long run. Linking this framework to the HCI requires a few additional steps. First, assume that the stock of human capital per worker that enters the production function, ​kh​ ​, is equal to the human capital of the average worker. Second, the human capital of the next generation, as measured in the HCI, and the human capital stock that enters the production func- tion need to be linked. This can be done by considering different scenarios. Imagine first a status quo scenario in which the expected years of quality-adjusted school and health as measured in the HCI today persist into the future. Over time, new entrants to the workforce with status quo health and education will replace current members of 8 For more details, see Kraay (2018). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 137 the workforce until eventually the entire workforce of the future has the expected years of quality-adjusted school and health captured in the current human capital index. Let​ k​h,NG​ = ​e​ϕ​s​ NG ​​ denote the future human capital stock in this baseline scenario, where​​ ​+γ​z​NG​ s​NG​ represents the number of quality-adjusted years of school of the next generation of workers, and ​γ ​z​NG​​is shorthand notation for the contribution of the two health indica- tors to productivity in the HCI in equation (4). Contrast this with a scenario in which the entire future workforce benefits from complete education and enjoys full health, resulting in a higher human capital stock, ​kh​*​ = e​ ​ϕ​s​ ​+γ​z​ ​​​, where ​​s​*​represents the benchmark * * of 14 years of high-quality school, and ​z​*​represents the benchmark of complete health. Assuming that total factor productivity and the physical capital-to-output ratio are the same in the two scenarios, the eventual steady-state GDP per worker in the two scenar- ios is as follows: _ _ ​−​s​*​)​+γ​​(​zNG ​ ​−​z​*​)​   (7) ​ ​ = ​​k​​ ​ = ​e​ ​y ​ ​k​ ​ h,NG ϕ​(​s​ NG ​y​*​ * h This expression is the same as the human capital index in equations (1)–(4) except for the term corresponding to survival to age 5 (because children who do not survive do not become part of the future workforce). This creates a close link between the human capital index and potential future growth. Disregarding the contribution of the survival probability to the HCI, equation (7) shows that a country with an HCI equal to ​x​ could achieve GDP per worker that would be ​1 / x​ times higher in the future if citizens enjoy complete education and full health (corresponding to ​x = 1​). For example, a country such as Morocco with an HCI value of around 0.5 could, in the long run, have future GDP per worker in this scenario of complete education and full health that is ​​_ 0.5​ = 2​ 1 times higher than GDP per worker in the status quo scenario. What this means in terms of average annual growth rates depends on how long the long run is. For example, under the assumption that it takes 50 years for these scenarios to materialize, then a doubling of future per capita income relative to the status quo corresponds to roughly 1.4 percentage points of additional growth per year. The calibrated relationship between the HCI and future income described here is sim- ple because it focuses only on steady-state comparisons. In related work, Collin and Weil (2018) elaborate on this by developing a calibrated growth model that traces out the dynamics of adjustment to the steady state. They use this model to trace out trajec- tories for per capita GDP and for poverty measures for individual countries and global aggregates under alternative assumptions for the future path of human capital. They also calculate the equivalent increase in investment rates in physical capital that would be required to deliver the same increases in output associated with improvements in human capital. 138 A ppendix A: T he H u man Capita l Index : Metho d o lo gy Figure A.1: Components of the HCI 1 600 Nicaragua Probability of Survival to Age 5, circa 2020 Kyrgyz Republic Solomon Islands Tuvalu Vanuatu Estonia Harmonized Test Scores, circa 2020 Rwanda Poland Vietnam Malawi 0.95 500 Burundi Ukraine Uzbekistan Kenya Mauritania Angola Cambodia Côte d’Ivoire Qatar Burundi Mali Benin Saudi Arabia 0.9 400 Guinea Kuwait Panama Chad Nigeria Dominican Republic South Africa Nigeria Ghana 0.85 300 6 8 10 12 6 8 10 12 Log GDP per capita at PPP, circa 2020 Log GDP per capita at PPP, circa 2020 14 1 Fraction of Children Under 5 Not Stunted, circa 2020 St. Lucia North Macedonia Samoa Turkey Kyrgyz Republic West Bank and Gaza Bulgaria Kazakhstan Expected Years of School, circa 2020 Nepal 12 Argentina Micronesia, Fed. Sts. Haiti Kiribati Zimbabwe Gambia, The Brunei Darussalam 0.8 Haiti Malaysia 10 Malawi Congo, Dem. Rep. Angola Gabon Botswana 8 Angola 0.6 Sudan Iraq Eswatini Timor−Leste Guatemala 6 Papua New Guinea Mali 4 Liberia 0.4 6 8 10 12 6 8 10 12 Log GDP per capita at PPP, circa 2020 Log GDP per capita at PPP, circa 2020 1 Morocco Nauru .9 West Bank and Gaza Adult Survival Rate, circa 2020 Vanuatu Tajikistan Solomon Islands Nepal Timor−Leste .8 Namibia .7 South Africa Zimbabwe Côte d’Ivoire Nigeria .6 Eswatini Central African Republic Lesotho .5 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure reports the most recent cross-section of 174 economies for the five HCI components (child survival, expected years of school, harmonized test scores, fraction of children under 5 not stunted, and adult survival), as used to calculate the 2020 HCI. Each panel plots the country-level averages for each component on the vertical and GDP per capita in PPP on the horizontal axis. The dashed line illustrates the fitted regression line between GDP per capita and the respective component. Scatter points above (below) the fitted regression line illustrate economies that perform higher (lower) in the outcome variable than their level of GDP would predict. Countries above the 95th and below the 5th percentile in distance to the fitted regression line are labeled. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 139 Figure A.2: Human Capital Index with uncertainty intervals 1 0.8 Finland Macao SAR, China Productivity Relative to Benchmark Cyprus Switzerland Italy Greece Russian Federation Slovak Republic Albania Bulgaria 0.6 Ecuador Azerbaijan Panama Zimbabwe Togo 0.4 Lesotho Côte d’Ivoire 0.2 6 8 10 12 Log Real GDP Per Capita Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the HCI (represented by a dot) and the uncertainty interval of the HCI for each economy. Appendix B: Back-calculated HCI T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 141 T he first iteration of the Human Capital Index, the 2018 HCI, made use of the best and most recently available data as of 2018. It was calculated for 157 econ- omies. As is common with indices and indicators, comprehensive revisions of the source data were done. For example, GDP series are revised quite often; even inter- national poverty numbers are revised as improved harmonization of survey data is implemented.9 The reasons for revisions are often to ensure temporal comparability and to incorporate the most accurate data available. In the case of the HCI, the index makes use of data from different institutions, and most of these institutions release their data annually. What this means is that the index value for 2018 will change, because the newest release of the data has slightly different num- bers. In the case of education, the changes are considerably more pronounced, because not only are the underlying data different, but in many cases data that are closer to 2018 are now available. When these elements are assembled to create the back-calcu- lated HCI for 2018, an index that is slightly different from the one published in 2018 emerges (figure B.1). On the y-axis, the countries for which it is now possible to obtain an index in 2018 are shown. These are countries where harmonized test scores are now available for the calculation of the index. Consequently, the number of economies with a back-calculated 2018 HCI is 167, 10 more than for the 2018 HCI. The countries are quite close to the 45-degree line, with some outliers like Tuvalu, where the 2018 HCI makes use of stunting as a proxy for health, and the back-calculated 2018 HCI makes use of adult mortality.10 The back-calculated global average for the 2018 index is 0.565 as opposed to 0.567 for the 2018 HCI. Of course, when looking at individual components, the differences between the 2018 HCI and the back-calculated 2018 HCI are starker. This is particularly relevant for the expected years of school (EYS). With a newer vintage of UNESCO Institute for Statistics (UIS) data, many of the enrollment rates used in the previous round of the HCI have been updated, leading to small changes, so data are added for years that are closer to 2018. This is further complicated because for some economies a preferred rate is now available (TNER>ANER>NER>GER).11 For example, in most economies where EYS 9 See Atamanov et al. (2019) for an example. 10 Stunting rates used in the 2018 HCI corresponded to a 2007 survey. The back-calculated 2018 HCI uses more recent adult mortality rates from 2012, from the World Health Organization, that were not previously available. 11 For details see the description of the construction of the EYS variable in appendix C. 142 A ppendix B: Bac k-ca lcu lated H C I Figure B.1: Comparing HCI 2018 and the back-calculated HCI 2018 1 0.8 Kazakhstan Germany United States Azerbaijan Seychelles HCI 2018 back−calculated 0.6 St. Lucia Palau El Salvador St. Vincent and the Grenadines Samoa India Papua New Guinea Tuvalu Marshall Islands 0.4 Eswatini 0.2 0 0 .2 .4 .6 .8 1 HCI 2018 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index and the 2018 Human Capital Index. Notes: Countries not present in the 2018 HCI, but present in the back-calculated HCI on the vertical axis, are Antigua and Barbuda, Dominica, Micronesia, Fed. Sts., Grenada, St. Kitts and Nevis, St. Lucia, Republic of the Marshall Islands, Palau, St. Vincent and the Grenadines, and Samoa. increased by at least half a year, the data come from a year that is closer to 2018, and in many cases there is a move to a preferred enrollment type. The back-calculated HCI makes use of the better and more recent data available. This allows for an index that better reflects the human capital that a child born in that year could aspire to achieve. The back-calculated HCI 2018 scores by country are reported in Table C2.3. Appendix C: HCI Component Data Notes T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 145 1. UNDER-5 MORTALITY RATES The probability of survival to age 5 is calculated as the complement of the under-5 mortality rate. The under-5 mortality rate is the probability of a child born in a speci- fied year dying before reaching the age of 5 if subject to current age-specific mortality rates. It is frequently expressed as a rate per 1,000 live births, in which case it must be divided by 1,000 to obtain the probability of dying before age 5. Under-5 mortality rates are calculated by the United Nations Interagency Group for Child Mortality Estimation (IGME) based on mortality as recorded in household sur- veys and vital registries. The IGME compiles and assesses the quality of all available nationally representative data relevant to the estimation of child mortality, including data from vital registration systems, population censuses, household surveys, and sam- ple registration systems. Globally, birth registration coverage remains inadequate, hav- ing increased by only 7 percentage points (from 58 percent to 65 percent) in the past decade.12 In Sub-Saharan Africa, only eight countries have coverage of 80 percent or more for under-5 birth registration.13 The IGME assesses data quality, recalculates data inputs and makes adjustments if needed by applying standard methods. It then fits a statistical model to these data to generate a smooth trend curve that averages over possibly disparate estimates from the different data sources for a country and, finally, it extrapolates the model to a target year. Data are reported annually and cover 198 countries. The IGME estimates are disaggregated by gen- der and include uncertainty intervals corresponding to 95 percent confidence intervals. 2020 Update Under-5 mortality rates for the 2020 update of the HCI come from the September 2019 update of the IGME estimates, available at the Child Mortality Estimates website, http:// www.childmortality.org/.14 Under-5 mortality rates for the 2020 HCI come from 2019, while data for the back- calculated 2018 HCI come from 2017. Data for the baseline comparator year of 2010 12 UNICEF, A Passport to Protection: A Guide to Birth Registration Programming (New York: UNICEF, 2013). 13 United Nations Statistics Division, December 2014, “Coverage of Birth and Death Registration,” http://un- stats.un.org/unsd/demographic/CRVS/CR_coverage.htm. 14 United Nations Interagency Group for Child Mortality Estimation (UNIGME, 2019). “Levels and Trends in Child Mortality, Report 2019,” United Nations Children’s Fund, New York. 146 A ppendix C : H C I C omp onent Data N otes Figure C1.1: Comparing original and back-calculated 2018 under-5 mortality rates 0.15 Under−5 Mortality Rates, 2018 back−calculated Nigeria Guinea 0.1 Madagascar 0.05 Lao PDR 0 0 .05 .1 .15 Under−5 Mortality Rates, 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the under-5 mortality rates as used in the HCI of 2018 (on the horizontal axis), and the under-5 mortality rates used for the back-calculated HCI of 2018 (on the vertical axis). The figure indicates differences that arise due to data updates. come from 2010. Since under-5 mortality rates are estimated by modeling all available child mortality data from vital registration systems, population censuses, household surveys, and sample registration systems, every new release of data from the IGME updates estimates for all the years in the time series. As a result, data for the same past year might differ slightly across updates. Values for under-5 mortality rates used to produce the back-calculated HCI 2018 are aligned with but not the same as those used in the previous iteration of the HCI, as illustrated in figure C1.1. Data from the two vintages align along the 45-degree line. The figure highlights the four countries where under-5 mortality rates have changed by more than 10 deaths per 1,000 live births or more. The largest revisions were for Nigeria (which went from 100 to 122 deaths per 1,000 live births) and Guinea (which went from 86 to 103 deaths per 1,000 live births). Figure C1.2 reports the most recent cross-section of under-5 mortality rates used to calculate the 2020 HCI. Child mortality rates range from around 0.002 (2 per 1,000 live births) in the richest countries to around 0.120 (120 per 1,000 live births) in the poorest countries. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 147 Figure C1.2: Under-5 mortality rates, HCI 2020 0.15 Chad Under−5 Mortality Rate, circa 2020 0.1 Ethiopia Congo, Rep. 0.05 India Senegal South Africa Indonesia Morocco Iran, Islamic Rep. Seychelles Kuwait Ecuador Thailand Oman China Bulgaria Switzerland Greece Hungary Cyprus 0 Finland Ireland 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots under-5 mortality rates (on the vertical axis) against log GDP per capita at 2011 USD PPP (on the horizontal axis). Figure C1.3: Sex-disaggregated under-5 mortality rates 0.15 Sex-disaggregated Under−5 Mortality Rate, circa 2020 Chad 0.1 Ethiopia 0.05 Congo, Rep. Senegal India South Africa Indonesia Morocco Seychelles Iran, Islamic Rep. Ecuador Thailand Oman China Bulgaria Kuwait Greece Hungary Cyprus Switzerland Ireland 0 Finland 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots sex-disaggregated under-5 mortality rates. The solid dot indicates the national average, the triangle is used to show the average value for girls, and the horizontal line shows the average value for boys. 148 A ppendix C : H C I C omp onent Data N otes Figure C1.4: Under-5 mortality rates by income group and region Low income 0.07 Lower-middle income 0.04 Upper-middle income 0.02 High income 0.01 0 .02 .04 .06 .08 Under−5 Mortality Rate, circa 2020 Sub−Saharan Africa 0.07 South Asia 0.04 East Asia & Pacific 0.02 Latin America & Caribbean 0.02 Middle East & North Africa 0.02 Europe & Central Asia 0.01 North America 0.01 0 .02 .04 .06 .08 Under−5 Mortality Rate, circa 2020 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot regional and income group average values for under-5 mortality rates. Under-5 mortality rates tend to be slightly lower for girls than for boys, as reported in figure C1.3. In the figure, the solid dot indicates the country average, the triangle indicates the average for girls, and the horizontal bar indicates the average for boys. The average under-5 mortality rate for boys was 0.03 (30 deaths per 1,000 live births), compared to 0.025 for girls. Figure C1.4 reports average child mortality rates by income group and by World Bank region. Mortality rates tend to be highest in low-income countries, and regional aver- ages are highest in Sub-Saharan Africa and South Asia, reflecting that poor countries continue to bear a disproportionate burden of child mortality. 2. EXPECTED YEARS OF SCHOOL15 The expected years of school (EYS) component of the HCI captures the number of years of school a child born today can expect to achieve by age 18, given the prevail- ing pattern of enrollment rates in her country. Conceptually, EYS is simply the sum 15 This section borrows heavily from the Technical Appendix of Kraay (2018). T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 149 of enrollment rates by age from age 4 to 17. Because age-specific enrollment rates are not broadly or systematically available, more readily-available data on enrollment rates by level of school are used to approximate enrollment rates in different age brackets. Pre-primary enrollment rates approximate the enrollment rates for 4- and 5-year-olds, primary enrollment rates approximate for 6- to 11-year-olds, lower-secondary rates approximate for 12- to 14-year-olds, and upper-secondary rates approximate for 15 -to 17-year-olds. Cross-country definitions in school starting ages and the duration of the various levels of school imply that these will only be approximations of the number of years of school a child can expect to complete by age 18. Given that the objective is to obtain a close proxy to age-specific enrollment rates, the pre- ferred measure is the “total net enrollment rate” (TNER). TNER measures the fraction of chil- dren in the theoretical age range for a given level of school who are in school at any level. For many countries, the TNER is not readily available for all levels and thus, in many instances, less preferred rates are used. The order of preference for the use of enrollment rates is: 1. Total net enrollment rates (TNER): TNER measures the fraction of children in the theoretical age range for a given level of school who are in school at any level. For pre-primary, because there is no level before pre-primary, TNER is not available, and ANER is the preferred measure. 2. Adjusted net enrollment rates (ANER): ANER measures the fraction of children in the theoretical age range for a given level of school who are in that level or the level above. 3. Net enrollment rates (NER): NER measures the fraction of children in the theoreti- cal age range for a given level of school who are in that level of school. 4. Gross enrollment rates (GER): GER measures the number of children of any age who are enrolled in that given level as a fraction of the number of children in that age range. The conceptually appropriate enrollment rate to approximate enrollment rates by age brackets is the repetition-adjusted total net enrollment rate. The primary source for enrollment and repetition rates is the United Nations Educational, Scientific, and Cultural Organization’s Institute for Statistics (UIS),16 revised and supplemented with data provided by World Bank country teams that participated in an extensive data review process. In cases where the resulting data on total net enrollment rates are incomplete, adjusted net enrollment rates, net enrollment rates, or gross enrollment rates are used instead in that order of priority. The same enrollment rate type is used for a given level of education over time. Because expected years of school is constructed based primarily on administrative data on enrollment rates, uncertainty intervals are not available for this component of the HCI. This does not imply that there is no measurement error, but because it comes 16 The main source for enrollment data from UIS is administrative data. Data are collected by UIS on an an- nual basis from official national statistical authorities. The data are released in September of every year and include national data for the school or reference year ending in the previous year. The national data are then updated in February, which completes the UIS publication of educational data for the data collection effort of the previous reference year (UIS 2018). 150 A ppendix C : H C I C omp onent Data N otes from administrative data there is no error due to modeling or sampling.17 Consequently, uncertainty in the measurement of expected years of school is not reflected in the uncertainty intervals of the overall HCI. EYS is calculated as follows: ​EYS = ​∑ 4i​rat​ei​​​Yi​​ ; i = pre-primary, primary​_​, lower-secondary, upper-secondary​      (1) where ​rat​e​i​​ is the enrollment rate for the preferred enrollment type available for that level, and ​Yi​​ is the number of years corresponding to each level.18 Enrollment rates for 2020 and 2010 Temporal coverage for enrollment rates is not complete in the UIS public database. Consequently, the first step toward ensuring that the rates used are the most recent and accurate relies on getting inputs from World Bank specialists working on each country to validate and provide more recent values when available.19 Enrollment rates for 2020 for each school level and for the four enrollment rate types (TNER, ANER, NER, GER) are obtained from UIS.20 Any inputs from World Bank teams working on specific countries are then added to the corresponding enrollment rates. Existing gaps for 2019 in enrollment rates for each level and country are filled by setting the 2019 enrollment rate equal to the latest enrollment rate available for that enroll- ment rate type. This is referred to as the “carryforward” rule. The rule is applied if the latest available enrollment rate is not older than 10 years.21 This process ensures that the HCI of 2020 and the back-calculated HCI for 2018 are done in a similar way to the first version of the HCI released in 2018. Additionally, enrollment rates are adjusted for repetition, where repetition rates are available, otherwise a repetition rate of 0 is assumed. Finally, enrollment rate types are chosen based on the filled series (that is using the rates for 2019 where gaps have been filled) and based on the following order of preference: TNER, ANER, NER, and GER.22 In the current HCI update, an effort has been made to also populate an HCI for 2010 using data circa 2010 in addition to data circa 2020.23 Since data collection and avail- ability generally improve over time, enrollment rates for 2010 and older are less likely 17 An important agenda concerns the frequent and substantial discrepancies between household survey–based measures of school enrollment and administrative records. D’Souza, Gatti, and Kraay (2019) discuss these briefly. 18 Y = 2 for pre-primary, Y = 6 for primary, Y= 3 for lower-secondary, and Y = 3 for upper-secondary. 19 For the 2020 update this process was conducted between January 29 and April 29, resulting in revised en- rollment rates for all levels, which are available in individual country files on https://www.worldbank.org/en/ publication/human-capital. 20 http://data.uis.unesco.org/ 21 The exceptions to this rule are for Fiji, Kiribati, and Kenya, where the most recent data available are pre-2010. 22 Note that one level of schooling may use TNER while another uses NER. However, for a given level of edu- cation, the same enrollment type is used over time. 23 This is done for all countries where the same test is available in or close to the specified year. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 151 to be available than more recent rates. This means that the rule from the first edition of the HCI used to obtain an EYS measure for 2020 and 2018 cannot be applied to obtain rates for 2010, because it is not possible to apply the carryforward rule for all econo- mies where comparable data over time for other components of the HCI are available circa 2010 and circa 2020. Moreover, to allow that (a) the preferred enrollment type is used for 2020 and that (b) the enrollment rate type for a given grade in a given country is the same over time, a different rule is applied to fill in the year 2010 to ensure com- parability over time, to maximize country coverage. The rule used to produce EYS in 2010 relies on using annualized growth rates and is implemented as follows for each school level (i.e., pre-primary, primary, lower-second- ary, and upper-secondary): 1. For the chosen enrollment rate used in 2020, if a 2010 value from the same rate is available, and this is not the same rate as the one used to fill in 2020, then that value is used.24 2. If 2010 is still missing, then use the same enrollment type to obtain an annualized growth rate between the latest year with non-missing rate before 2010 and the ear- liest (closest to 2010) year with non-missing rate post 2010. Then apply the annu- alized growth rate (agr) to the rate of the most recent year before 2010 (year) with non-missing rate to obtain the rate for 2010: ​ ​ ​ = rat​eyear rat​e2010 ​ ​​(1 + agr)​2010−year​  (2) 3. If there is no rate available for the chosen enrollment type before 2010, then annu- alized growth rates are obtained using GER enrollment rates, which are available for most countries. Annualized growth rates are obtained between the latest year with non-missing rates for the year 2010 or before, if 2010 is not available, and the GER enrollment rate for the same year or after that of the earliest rate post-2010 of the preferred rate. For example, if the preferred rate is a TNER and its earli- est value is for 2012, then the annualized growth rate is obtained from GER. The annualized growth rate from GER is obtained between the first rate available on or before 2010, and the first year available after or equal to 2012, since the TNER is available on that year. This rate is then applied backwards to the TNER of 2012 to obtain a value for 2010. ​ ​ ​ = rat​eyear rat​e2010 ​ ​​(1+ag​rGER ​ )​ ​2010−year​​ (3) The process described above yields a value for expected years of school in 2010 for 99 out of 114 eligible economies. For the remaining 15 economies, exceptions to the rules detailed above are determined on a case-by-case basis in order to populate enrollment rates in 2010. 24 Exceptions are Qatar pre-primary and primary where the same value of 2010 is used. 152 A ppendix C : H C I C omp onent Data N otes Gender disaggregation Gender disaggregation is an important feature of the Human Capital Index. Although the rules presented in the previous section are meant to complete the EYS for both sexes, there are still adjustments required to ensure that EYS values for boys and girls are plausible. These adjustments are necessary because, although a certain enrollment type may be available for both sexes, it may lack sex-disaggregated information. In other instances, it may be necessary to adjust the disaggregated series because both val- ues for each gender are above (or below) those of the combined enrollment rate. To fill in the sex-disaggregated enrollment rates the following rules are applied: 1. For every year where rates for both genders and the aggregate are available, the male-to-female ratio and the population share of males and females are calculated. 2. For years that are missing a sex-disaggregated rate, the shares and ratios calculated in step (1) from the closest year available in the past (but not more than 10 years back) are used to impute missing values. 3. For the remaining years where the disaggregated enrollment rates for the preferred enrollment type are still missing, the male and female shares and, where available, the male-to-female ratio from GER enrollment rates are used to impute a value. It is still possible that the rules above, when applied, return inconsistent values, and it is necessary to adjust the disaggregated series when the male and female rates are both larger (or smaller) than the aggregate enrollment rate. In those cases, we adjust the dis- ​ ​​at the same aggregated enrollment rate to the value that leaves the aggregate rate ​rat e​ mf distance from each of the disaggregated rates. ​ ​+_ rat​ef​​− rat​em​ ​ ​ at​ef​*​​ (i) r = rat​emf ​ 2​ ​ ​+_ rat​em​ ​− rat​ef​​ (ii)​rat​em​* ​ = rat​emf ​ 2​ 2018 Back-calculated EYS Data for the 2020 update of EYS rely on data from UIS. UIS releases data in September of each year, and the release is completed in February of the following year. The 2020 February release of enrollment data from UIS is used for the update of EYS. The latest data release from UIS is complemented with rates obtained by World Bank staff.25 The updated data provide an opportunity to update EYS values from the 2018 vintage of the HCI to the latest information available to arrive at a back-calculated EYS for 2018. Because the update allows for the calculation of expected years of school where the data may be newer or come from a different enrollment type than what was used in the first vintage of the HCI, the HCI and EYS of the first vintage of the HCI are not comparable to the current vintage of the HCI and EYS. 25 World Bank staff working in each country obtain these data from local government sources, for example, the Ministry of Education or National Statistics Office. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 153 Figure C2.1: Comparing original and back-calculated 2018 expected years of school 14 Expected Years of School, 2018 back−calculated Azerbaijan 12 Kenya Zimbabwe Nicaragua Papua New Guinea Tuvalu Nigeria South Africa Bangladesh 10 Lesotho Madagascar 8 Mauritania Eswatini 6 4 4 6 8 10 12 14 Expected Years of School, 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the expected years of school as used in the HCI of 2018 (on the horizontal axis), and the expected years of school used for the back-calculated HCI of 2018 (on the vertical axis). The figure indicates differences that arise due to data updates. Differences between the HCI values of 2018 for the 2020 update and the 2018 vintage values may be due to a combination of three factors: 1. Data are updated in UIS data vintage, or by World Bank staff. 2. Data on enrollment from a more recent year are now available. 3. Different enrollment types may be available. In some cases, it will be possible to move to a more preferred enrollment type, while in others it is necessary to rely on a less preferred enrollment type. The latter may be the case if UIS has removed the series or because the series is too old. The average absolute deviation between the back-calculated EYS and the one for 2018 is 0.3 years. However, the changes are considerable for many countries (see figure C2.1). Although the differences between vintages are considerable, these are mostly due to the fact that the EYS measure generated in this round relies on more preferred rates, and/or newer data. For the 2018 back-calculated EYS, the enrollment data for at least one of the levels for 131 economies comes from a more recent year.26 For 85 economies, the enrollment rates for all levels correspond to a more recent year. In 21 economies, for at least one of the levels, it 26 The flip side is that for 15 economies at least one of the enrollment rates used comes from an older year than was available in 2018’s EYS. This is mostly because UIS revises the series, and in some instances, years may be removed from the series. 154 A ppendix C : H C I C omp onent Data N otes is necessary to change to a less preferred series. This is mostly in cases where the series has been removed in the update of the data. On the other hand, in 20 countries, for at least one level it is possible to calculate EYS with a more preferred type of enrollment rate. Figure C2.2 and figure C2.3 present details for countries where EYS under the new data vintage has increased by at least half a year. In most economies where EYS increased by at least half a year there is a move to a data point that is closer to 2018. The exception is for Zimbabwe, where all the enrollment rates correspond to the same year and are for the same enrollment type. In the case of Zimbabwe, the difference is explained as due to the change in the vintage of UIS data. The biggest change for Zimbabwe is observed for primary, where the rate increased by almost 10 percentage points. A different case to that of Zimbabwe can be observed for Côte d’Ivoire, where every data point comes from a more recent year. However, in this case the difference is also complicated because the previous EYS was built with rates that did not come from UIS but were drawn from government sources by World Bank staff. In the case of Papua New Guinea, the change is due to two factors. Not only are more recent data used for all levels, but also in all but primary the data being used are from a preferred series (figure C2.3). These three countries illustrate the multiple sources for the potential mismatch between the EYS value produced in 2018 and the updated 2018 back-calculated EYS. Figure C2.4 and figure C2.5 present details for countries where EYS, under the back-cal- culated 2018 EYS, has decreased by at least half a year. Only in Tanzania is it necessary to move to an older rate, but it is to a preferable type (GER to TNER), and it is only one year older (figure C2.4 for year, and figure C2.5 for types). In Bangladesh, the change in EYS is mostly driven by changes in pre-primary and lower-secondary. For pre-pri- mary, the back-calculated rate relies on data from 2017 versus 2011, although it is for a less preferred rate (GER versus ANER). Meanwhile, for lower-secondary the rate for the back-calculated EYS is for a more recent year but a less preferred rate (ANER versus TNER). In this case, it is necessary to move to a less preferred rate because the TNER series is no longer available in the UIS data vintage for years after 2010. In India, because the latest available TNER series in UIS is for 2013, World Bank staff has sourced more recent data. EYS is now built with age-specific enrollment profiles that make use of information from UDISE+ from the Ministry of Human Resource Development, as well as early childhood care and education enrollment from the Ministry of Women and Child Development and Entrepreneurship and population projections from the Ministry of Health and Family Welfare. The resulting EYS for the back-calculated HCI is 10.8 versus 10.2, which was used in the calculation for the 2018 HCI. Figure C2.4 and figure C2.5 should present enough evidence against the comparison of the 2018 HCI published in 2018 and the update of the HCI in 2020. For a more detailed look into the differences, table C2.1 presents enrollment data for all the countries where the absolute EYS change between the back-calculated 2018 and the 2018 versions of the index is greater than half a year. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 155 Figure C2.2: Vintage data year for back-calculated 2018 and 2018, where EYS increased by 0.5 years or more Pre−primary Primary Zimbabwe Zimbabwe West Bank and Gaza West Bank and Gaza Vietnam Vietnam Tonga Tonga Timor−Leste Timor−Leste South Africa South Africa Papua New Guinea Papua New Guinea Pakistan Pakistan Nigeria Nigeria Mauritania Mauritania Madagascar Madagascar Lesotho Lesotho Kenya Kenya India India Dominican Republic Dominican Republic Côte d’Ivoire Côte d’Ivoire Azerbaijan Azerbaijan 2005 2010 2015 2020 2005 2010 2015 2020 Lower−secondary Upper−secondary Zimbabwe Zimbabwe West Bank and Gaza West Bank and Gaza Vietnam Vietnam Tonga Tonga Timor−Leste Timor−Leste South Africa South Africa Papua New Guinea Papua New Guinea Pakistan Pakistan Nigeria Nigeria Mauritania Mauritania Madagascar Madagascar Lesotho Lesotho Kenya Kenya India India Dominican Republic Dominican Republic Côte d’Ivoire Côte d’Ivoire Azerbaijan Azerbaijan 2005 2010 2015 2020 2005 2010 2015 2020 2018 Back−calculated 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot the year of data used for calculation of EYS. Solid dots represent the data used for the back- calculated HCI of 2018, and the x’s indicate the data used for the calculation of the 2018 HCI. 156 A ppendix C : H C I C omp onent Data N otes Figure C2.3: Enrollment type for back-calculated 2018 and 2018, where EYS increased by 0.5 years or more Pre−primary Primary Zimbabwe Zimbabwe West Bank and Gaza West Bank and Gaza Vietnam Vietnam Tonga Tonga Timor−Leste Timor−Leste South Africa South Africa Papua New Guinea Papua New Guinea Pakistan Pakistan Nigeria Nigeria Mauritania Mauritania Madagascar Madagascar Lesotho Lesotho Kenya Kenya India India Dominican Republic Dominican Republic Côte d’Ivoire Côte d’Ivoire Azerbaijan Azerbaijan TNER ANER NER GER TNER ANER NER GER Lower−secondary Upper−secondary Zimbabwe Zimbabwe West Bank and Gaza West Bank and Gaza Vietnam Vietnam Tonga Tonga Timor−Leste Timor−Leste South Africa South Africa Papua New Guinea Papua New Guinea Pakistan Pakistan Nigeria Nigeria Mauritania Mauritania Madagascar Madagascar Lesotho Lesotho Kenya Kenya India India Dominican Republic Dominican Republic Côte d’Ivoire Côte d’Ivoire Azerbaijan Azerbaijan TNER ANER NER GER TNER ANER NER GER 2018 Back−calculated 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot the enrollment type used for calculation of EYS. Solid dots represent the data used for the back-calculated HCI of 2018, and the x’s indicate the data used for the calculation of the 2018 HCI. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 157 Figure C2.4: Vintage data year for back-calculated 2018 and 2018, where EYS decreased by 0.5 years or more Pre−primary Primary Vanuatu Vanuatu Tuvalu Tuvalu Tanzania Tanzania Solomon Islands Solomon Islands Seychelles Seychelles Panam Panam Nicaragua Nicaragua Germany Germany Eswatini Eswatini Bulgaria Bulgaria Bangladesh Bangladesh 2005 2010 2015 2020 2005 2010 2015 2020 Lower−secondary Upper−secondary Vanuatu Vanuatu Tuvalu Tuvalu Tanzania Tanzania Solomon Islands Solomon Islands Seychelles Seychelles Panam Panam Nicaragua Nicaragua Germany Germany Eswatini Eswatini Bulgaria Bulgaria Bangladesh Bangladesh 2005 2010 2015 2020 2005 2010 2015 2020 2018 Back−calculated 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot the year of data used for calculation of EYS. Solid dots represent the data used for the back- calculated HCI of 2018, and the x’s indicate the data used for the calculation of the 2018 HCI. 158 A ppendix C : H C I C omp onent Data N otes Figure C2.5: Enrollment type for back-calculated 2018 and 2018, where EYS decreased by 0.5 years or more Pre−primary Primary Vanuatu Vanuatu Tuvalu Tuvalu Tanzania Tanzania Solomon Islands Solomon Islands Seychelles Seychelles Panama Panama Nicaragua Nicaragua Germany Germany Eswatini Eswatini Bulgaria Bulgaria Bangladesh Bangladesh TNER ANER NER GER TNER ANER NER GER Lower−secondary Upper−secondary Vanuatu Vanuatu Tuvalu Tuvalu Tanzania Tanzania Solomon Islands Solomon Islands Seychelles Seychelles Panama Panama Nicaragua Nicaragua Germany Germany Eswatini Eswatini Bulgaria Bulgaria Bangladesh Bangladesh TNER ANER NER GER TNER ANER NER GER 2018 Back−calculated 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot the enrollment type used for calculation of EYS. Solid dots represent the data used for the back-calculated HCI of 2018, and the x’s indicate the data used for the calculation of the 2018 HCI. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 159 Figure C2.6: EYS 2020 and log GDP per capita at PPP 15 Finland Ireland Cyprus China Greece Switzerland Ecuador Seychelles Hungary Thailand Oman Expected Years of School, circa 2020 Indonesia Bulgaria Kuwait Iran, Islamic Rep. India Morocco South Africa 10 Congo, Rep. Ethiopia Senegal Chad 5 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots expected years of school (on the vertical axis) against log GDP per capita at 2011 USD PPP (on the horizontal axis). 2020 Update The 2020 EYS shows a high rank correlation to the EYS from 2018, suggesting that the higher a country’s 2018 ranking in EYS, the higher the ranking in EYS in 2020 and vice versa. Also, there is a strong positive relationship between the 2020 EYS and log GDP per capita (figure C2.6). Expected years of school tend to be slightly higher for girls than for boys, as reported in figure C2.7. In the figure, the solid dot indicates the country average, the triangle indicates the average for girls, and the horizontal bar indicates the average for boys. The average expected years of school for boys was 11.3 compared to 11.4 for girls. Disparity in expected years of school is lower in richer countries. Figure C2.8 reports average expected years of school by income group and by World Bank region. Expected years of school tend to be lowest in low-income countries, and regional averages are lowest in Sub-Saharan Africa and South Asia. This suggests that much work remains to be done to close the gap in low-income countries. 160 A ppendix C : H C I C omp onent Data N otes Figure C2.7: Sex-disaggregated expected years of school 14 Ireland Finland Cyprus China Greece Switzerland Seychelles Hungary Sex-disaggregated Expected Years of School, circa 2020 Ecuador Thailand Oman Indonesia Bulgaria 12 Iran, Islamic Rep. Kuwait India Morocco South Africa 10 Congo, Rep. 8 Ethiopia Senegal 6 Chad 4 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots sex-disaggregated expected years of school. The solid dot indicates the national average, the triangle is used to show the average value for girls, and the horizontal line shows the average value for boys. Figure C2.8: Expected years of school by income group and region Low income 7.6 Lower-middle income 10.4 Upper-middle income 11.8 High income 13.2 0 5 10 15 Expected Years of School, circa 2020 Sub−Saharan Africa 8.3 South Asia 10.8 Middle East & North Africa 11. 6 East Asia & Pacific 11.9 Latin America & Caribbean 12. 1 Europe & Central Asia 13. 1 North America 13. 3 0 5 10 15 Expected Years of School, circa 2020 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot regional and income group average values for expected years of school. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 161 3. HARMONIZED TEST SCORES The school quality adjustment is based on a large-scale effort to harmonize interna- tional student achievement tests from several multicountry testing programs to pro- duce the Global Dataset on Education Quality. A detailed description of the test score harmonization exercise is provided in Patrinos and Angrist (2018), and the HCI draws on an updated version of this dataset as of January 2020.27 The dataset harmonizes scores from three major international testing programs—the Trends in International Mathematics and Science Study (TIMSS) program, the Progress in International Reading Literacy Study (PIRLS), and the Programme for International Student Assessment (PISA)—as well as three major regional testing programs—the Southern and Eastern Africa Consortium for Monitoring Educational Quality (SACMEQ), the Program for the Analysis of Education Systems (PASEC), the Latin American Laboratory for Assessment of the Quality of Education (LLECE), and the Pacific Island Learning and Numeracy Assessment (PILNA). It also incorporates Early Grade Reading Assessments (EGRAs) coordinated by the United States Agency for International Development. The harmonization methodology relies on the production of an “exchange rate” between international student achievement tests and their regional counterparts, which can then be used to place tests on a common scale. Test scores are converted into TIMSS units as the numeraire, corresponding roughly to a mean of 500 and a standard deviation across students of 100 points. The exchange rate is based on the ratio of average country scores in each program to the corresponding country scores in the numeraire testing program for the set of countries participating in both the numeraire and the other testing program. For example, consider the set of countries that participate in both the PISA and the TIMSS assessments. The ratio of average PISA scores to average TIMSS scores for this set of countries provides a conversion factor for PISA into TIMSS scores that can then be used to convert the PISA scores of all countries into TIMSS scores. The exchange rate is calculated pooling all overlapping observations between 2000 and 2017 and is therefore constant over time. This ensures that within-country fluctuations in harmonized test scores over time for a given test- ing program reflect only changes in the test scores themselves and not changes in the conversion factor between tests.28 The most recent update of the dataset also uses the 2000–17 period to calculate exchange rates, so that the rates between testing programs do not change between the 2018 and 2020 versions of the database. 2020 Update The 2020 update of the Global Dataset on Education Quality extends the database to 184 countries and economies, drawing on a large-scale effort by the World Bank to col- lect learning data globally. 27 For the latest updates on the HTS see Angrist et al. (2019). 28 The one exception to this is the 2007 and 2014 PASEC rounds, which were not designed to be intertempo- rally comparable and in which different overlapping countries were used to construct the exchange rate in the two periods. 162 A ppendix C : H C I C omp onent Data N otes Updates to the database come from new data from PISA 2018, PISA for Development (PISA-D), PILNA, and EGRA. The database adds 20 new countries (8 using EGRAs, 8 using PILNA, 3 using PISA and PISA-D, and 1 using a national TIMSS-equivalent assess- ment). This brings the percentage of the global school-age population represented by the database to 98.7 percent. In addition, more recent data points have been added for 94 countries (75 from PISA 2018, 7 from PISA-D, 5 from EGRAs, and 7 from PILNA). In most cases, the tests are designed to be nationally representative. There are, however, some notable cases in which they are not. In the case of China, extrapolations are needed to arrive at nationally representative estimates, since only a small number of relatively affluent regions have participated in PISA assessments. For India, the only internation- ally comparable assessment is the 2009 PISA. Instead, recent national assessment data and exchange rates with international benchmarks derived from the UNESCO Institute for Statistics (UIS) Global Alliance to Monitor Learning (GAML) process are used to esti- mate a national harmonized test score (HTS).29 In a number of countries, EGRAs are not nationally representative and are identified as EGRANR in the data documentation.30 In cases where countries participate in multiple testing programs, a hierarchy of tests is applied to determine which HTS to use. This hierarchy is based on the strength of the underlying test construction, the number of overlapping countries to produce the exchange rate, and consistency in administration, procedures, and documentation over time. The first HTS choice is an international test like the PISA, TIMSS, or PIRLS. The next-choice HTS is a regional test, like LLECE, SACMEQ, PASEC, and PILNA (in that order). Finally, if neither an international nor a regional test is available, a country is assigned an HTS that comes from an EGRA. The one exception to this rule is Yemen, where TIMSS data from 2007 and 2011 yield implausibly low scores and are replaced with EGRA data from 2011. Uncertainty intervals for HTSs are constructed by bootstrapping. Patrinos and Angrist (2018) take 1,000 random draws from the distribution of subject-grade average test scores for each test in their dataset. They then form exchange rates and calculate HTSs in each bootstrapped sample. The 2.5th and 97.5th percentiles of the distribution of the resulting HTSs across bootstrapped samples constitute the lower and upper bounds of the uncertainty interval for the HTS. Test scores are harmonized by subject and grade and are then averaged across subjects and grades.31 29 For China, the extrapolations are based on the relationship between available internationally comparable test scores and per capita income levels in the regions where these tests were administered, updating the calculations in Annex 4 of Patrinos and Angrist (2018) to include the newly available data from the 2018 PISA round. For India, the methodology takes advantage of a process coordinated by UIS to define common “basic minimum proficiency (BMP)” thresholds across different national and international assessments, including India’s national assessment carried out in 2017. This process creates an equivalence between a BMP threshold in India’s national assessment and the corresponding value in TIMSS/PIRLS. This information is used to rescale the mean scores in India’s national assessment into internationally comparable HTS units. Detailed notes on the methodology and calculations for the HTS for China and India are available on request. 30 For the 2020 HCI, 13 economies have an HTS that comes from a non representative EGRA: Bangladesh; Central African Republic; Congo, Dem. Rep.; Ethiopia; Iraq; Jamaica; Lao PDR; Liberia; Mali; Myanmar; Nigeria; Pakistan; and South Sudan. 31 See Patrinos and Angrist (2018), for further details. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 163 HTSs for the 2020 HCI come from the most recently available test as of 2019; while data for the back-calculated 2018 HCI come from the most recent test available as of 2017. Data for the baseline comparator year of 2010 are populated for each country using the test closest to 2010, typically with a minimum gap of five years between the test used to populate the 2010 and 2020 cross-sections. Some exceptions to this rule include Bahrain, Botswana, Islamic Republic of Iran, Kuwait, Oman, and South Africa, where data from the 2011 TIMSS or PIRLS are used to calculate the 2010 HCI, and data from the 2015 TIMSS or PIRLS are used to calculate the 2020 HCI. In addition, data for Timor-Leste come from a 2009 and 2011 EGRA, while data for Vietnam come from a 2012 and 2015 PISA for the 2010 HCI, and 2020 HCI respectively. In order to ensure the comparability of HTSs across time, we ensure that the 2010 and 2020 cross-sections are populated with scores that come from the same testing program. That is, if a country has an HTS from a PISA test circa 2020, it must also have scores from another PISA test circa 2010 to be included in the over-time comparison. The five excep- tions are Algeria, Morocco, North Macedonia, Saudi Arabia, and Ukraine. For Algeria, harmonized test scores from the PIRLS or the TIMSS in 2007 are used to populate the 2010 HCI, while harmonized test scores based on the PISA in 2015 are used to populate the 2020 HCI. For Morocco, North Macedonia, Saudi Arabia, and Ukraine, data from PIRLS or TIMSS in 2011 are used for the 2010 HCI, while data from PISA 2018 are used for the 2020 HCI. To maximize comparability with PISA, only scores from secondary level schooling are considered for these five countries for the 2010 HCI. Applying these rules yields a sample of 103 countries with test scores in both 2010 and 2020. Test scores used to produce the back-calculated HCI 2018 are similar to those used in the previous iteration of the HCI, as illustrated in figure C3.1. Data from the two vintages align almost perfectly along the 45-degree line. This is because outcomes for these coun- tries come from the same test and the same harmonization methodology. The figure also highlights the ten countries where test scores have changed because a more recent test was made available in the latest version of the database or, as in the case of China and India, because alternate methodologies were used to refine estimates of national average learning outcomes (see table C3.1 for details on changes in the source of test data). In the case of El Salvador, a choice guided by consultations with the country team was made to replace the previous test used (TIMSS/PIRLS from 2007) with a 2006 LLECE for reading to enhance comparability to the 2018 EGRA (not representative) used in 2020. Figure C3.2 reports the most recent cross-section of test scores used to calculate the 2020 HCI. HTSs range from around 575 in the richest countries to around 305 in the poorest countries. To interpret these units, note that 400 corresponds to the benchmark of “low proficiency” in TIMSS at the student level, while 625 corresponds to “advanced proficiency.” Test scores tend to be slightly higher for girls than for boys, as reported in figure C3.3. In the figure, the solid dot indicates the country average, the triangle indicates the aver- age for girls, and the horizontal bar indicates the average for boys. Globally, the average HTS for boys was 420, compared with 430 for girls. 164 A ppendix C : H C I C omp onent Data N otes Figure C3.1: Comparing original and back-calculated 2018 test scores 600 Harmonized Test Scores, 2018 back−calculated 500 El Salvador China 400 India Tonga Gambia, The Tuvalu Haiti Vanuatu Congo, Dem. Rep. Nigeria 300 300 400 500 600 Harmonized Test Scores, 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the harmonized test scores as used in the HCI of 2018 (on the horizontal axis), and the harmonized test scores used for the back-calculated HCI of 2018 (on the vertical axis). The figure indicates differences that arise due to data updates. Table C3.1: Source data for countries with different values in 2018 and back- calculated 2018 2018 VINTAGE 2020 VINTAGE COUNTRY Test Year Value Test Year Value China PISA/PIRLS (Extrapolated) 2015 456 PISA/PIRLS (Extrapolated) 2015 441 Congo, Dem. Rep. EGRANR 2012 318 EGRANR 2015 310 El Salvador TIMSS/PIRLS 2007 362 LLECE 2006 438 Gambia, The EGRA 2011 338 EGRA 2016 353 Haiti EGRANR 2013 345 EGRA 2016 339 India PISA 2009 355 NAS 2017 399 Malaysia TIMSS 2015 468 TIMSS/PIRLS 2015 468 Nigeria EGRANR 2010 325 EGRANR 2014 309 Tonga EGRA 2014 376 PILNA 2015 370 Tuvalu * EGRA 2016 387 EGRA 2016 351 Vanuatu EGRA 2010 356 PILNA 2015 332 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: Data for Tuvalu from the 2016 EGRA were revised once student-level data were made available to the HTS team. NAS = National Achievement Survey. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 165 Figure C3.2: Harmonized test scores, HCI 2020 600 Finland Harmonized Test Scores, circa 2020 Ireland Switzerland 500 Cyprus Hungary Greece Seychelles China Bulgaria Iran, Islamic Rep. Thailand Oman Ecuador Senegal 400 India Indonesia Morocco Kuwait Congo, Rep. Ethiopia South Africa Chad 300 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots harmonized test scores (on the vertical axis) against log GDP per capita at 2011 USD PPP (on the horizontal axis). Figure C3.3: Sex-disaggregated harmonized test scores 600 Sex Disaggregated Harmonized Test Scores, circa 2020 Finland Ireland Switzerland 500 Cyprus Hungary Greece Seychelles China Bulgaria Iran, Islamic Rep. Thailand Oman Ecuador Senegal 400 India Indonesia Morocco Kuwait Congo, Rep. Ethiopia South Africa Chad 300 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots sex-disaggregated harmonized test scores. The solid dot indicates the national average, the triangle is used to show the average value for girls, and the horizontal line shows the average value for boys. 166 A ppendix C : H C I C omp onent Data N otes Figure C3.4: Harmonized test scores by income group and region Low income 356 Lower-middle income 392 Upper-middle income 411 High income 487 0 100 200 300 400 500 Harmonized Test Scores, circa 2020 South Asia 374 Sub−Saharan Africa 374 Latin America & Caribbean 405 Middle East & North Africa 407 East Asia & Pacific 432 Europe & Central Asia 479 North America 523 0 100 200 300 400 500 Harmonized Test Scores, circa 2020 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot regional and income-group average values for harmonized test scores. Figure C3.4 reports average test scores by income group and by World Bank region. Test scores tend to be lowest in low-income countries, and regional averages are lowest in South Asia and Sub-Saharan Africa. 4. UNDER-5 STUNTING RATES The fraction of children under 5 not stunted is calculated as the complement of the under-5 stunting rate. The stunting rate is defined as the share of children under the age of 5 whose height is more than two reference standard deviations below the refer- ence median for their ages. The reference median and standard deviations are set by the World Health Organization (WHO) for normal healthy child development.32 Child- level stunting prevalence is averaged across the relevant 0–5 age range to arrive at an overall under-5 stunting rate. The stunting rate is used as a proxy for latent health of the population, in addition to the adult survival rate, in countries where stunting data are available, as discussed below. 32 World Health Organization (2009). The WHO Multicentre Growth Reference Study (MGRS). Geneva, World Health Organization. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 167 Data on stunting rates are taken from the Joint Child Malnutrition Estimates ( JME) database,33 managed by UNICEF, WHO, and the World Bank. The database reports the prevalence of stunting, wasting, and underweight, and is populated with estimates from survey data, gray literature, and reports from national authorities, reviewed by the JME interagency team. If required, data are reanalyzed to produce nationally representa- tive estimates for the appropriate age cohort (0–5 years), comparable across countries and across time. Surveys presenting anthropometric data for age groups other than 0–59 months or 0–60 months are adjusted using national survey results—gathered as close in time as possible—from the same country that include the age range 0–59/60 months. National rural estimates are adjusted similarly using another national survey for the same country as close in time as possible with available national urban and rural data to derive an “adjusted national estimate.” Historical data that use different growth reference standards are reanalyzed to produce estimates based on WHO stan- dards when raw data are available. If raw data are unavailable, estimates are converted to WHO-based prevalence using an algorithm developed by Yang and de Onis (2008).34 The JME reports stunting rates from surveys and administrative data and is updated twice a year, in March and September. The HCI team supplements stunting data from the JME with data provided by country teams for five countries: Bhutan, Chile, Fiji, Indonesia, and Timor-Leste. This is primarily to include more recent surveys that have not yet been incorporated in the JME. The March 2020 update of the JME reports data for 152 countries and 887 country-year observations. About 50 percent of the JME data comes from the Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys (MICS). Both are nation- ally representative household surveys that collect data on measures of population, health, and nutrition.35 About 10 percent of JME data comes from country nutrition surveillance programs, while the rest of the database is populated using national sur- veys that collect anthropometric data and measure stunting directly. The JME database reports sex-disaggregated stunting rates for 56 percent of the sur- veys. It also reports 95 percent confidence intervals around estimates of stunting for about 40 percent of the observations, primarily those on which the JME team had access to record-level survey data. Absent better alternatives, the HCI team imputes confidence intervals for the remaining observations in the JME database using the fitted values from a regression of the width of the confidence interval on the stunting rate. 33 See JME (UNICEF-WHO-World Bank Joint Child Malnutrition Estimates) (database), 2020 edition, United Nations Children’s Fund, New York, https:// data.unicef.org/resources/jme/. 34 United Nations Children’s Fund, World Health Organization, The World Bank (2012). UNICEFWHO-World Bank Joint Child Malnutrition Estimates. New York, UNICEF; Geneva, WHO; Washington, DC, The World Bank. 35 The DHS program has fielded over 400 surveys across 90 countries, while over 300 MICS have been carried out in more than 100 countries. 168 A ppendix C : H C I C omp onent Data N otes Surveys from low- and middle-income countries make up 90 percent of the JME database. High-income countries tend to have much lower average stunting rates (the national average for the 13 high-income countries in the JME sample is 6 percent) and are less likely to regularly monitor stunting through frequent surveys. However, some high-income countries like Kuwait, Oman, and the Unites States continue frequent monitoring of stunting prevalence through national surveys. Inconsistent measure- ment is of greater concern in middle- or low-income countries where stunting rates continue to be elevated. The most recent survey for 33 countries in the JME database is more than five years old, and it is around 10 years old for 10 countries. On the other hand, countries like Peru and Senegal elected to field DHS surveys annually. The con- tinuous DHS played a key role in Peru’s national strategy for early childhood devel- opment, Crecer, which helped reduce the country’s rate of chronic malnutrition from 28 percent in 2005 to 13 percent in 2016, with an even pace of change among rural and urban children.36 In the JME data, the average gap between surveys for countries with at least two surveys is 5.6 years, and five years when high-income countries are excluded. 2020 Update Stunting rates for the 2020 update of the HCI come from the March 2020 update of the JME database, available at the UNICEF website, https://www.who.int/publications-detail/jme- 2020-edition.37 Relative to the 2018 edition of the HCI, this latest update to the database allows us to update stunting rates for 54 countries, and add stunting rates for Argentina, Bulgaria, and Uzbekistan, which did not have a rate in the previous iteration of the HCI. Stunting rates for the 2020 HCI come from the most recently available survey as of 2019, while data for the back-calculated 2018 HCI come from the most recent sur- vey available as of 2017. Data for the baseline comparator year of 2010 are populated for each country using the survey closest to 2010 that was fielded between 2005 and 2015. When populating the 2010 cross-section, we ensure a minimum gap of five years between the survey used to populate the 2010 and 2020 cross-sections. To maximize the overlap among the three cross-sections, we do not rely on stunting rates in the cal- culation of the HCI for high-income countries, even when stunting data are available for some of these countries. This is because stunting rates typically come from surveys that are 5–10 years old for these countries. Further, to ensure consistency across time periods, we only use stunting data to calculate the HCI for a country if such data are available in both 2010 and 2020. This does not prevent the calculation of an HCI score for high-income countries or countries missing stunting in any period; we simply use the adult survival rate as the proxy for latent health in our calculations. Values for stunting rates used to produce the back-calculated HCI 2018 are very similar to those used in the previous iteration of the HCI, as illustrated in figure C4.1, where 36 Marini and Rokx (2017). 37 Joint Child Malnutrition Estimates ( JME): Levels and trends (2020 edition), database, https://data.unicef. org/resources/jme-report-2020. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 169 Figure C4.1: Comparing original and back-calculated 2018 stunting rates 0.6 Timor−Leste Stunting Rates, 2018 back−calculated 0.4 India Sierra Leone Togo Burkina Faso 0.2 Tajikistan Albania Mongolia 0 0 .2 .4 .6 Stunting Rates, 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots the stunting rates as used in the HCI of 2018 (on the horizontal axis), and the stunting rates used for the back-calculated HCI of 2018 (on the vertical axis). The figure indicates differences that arise due to data updates. data from the two vintages align almost perfectly along the 45-degree line. The figure highlights 8 countries where stunting rates have changed by 3 percentage points or more in the back-calculated HCI 2018 versus the original HCI 2018. This is predomi- nantly because the March 2020 update of the JME makes a more recent survey avail- able or, in the case of Djibouti and Sierra Leone, because JME estimates have been updated following a reanalysis of survey data (see table C4.1). Figure C4.2 reports the most recent cross-section of stunting rates used to calculate the 2020 HCI. Stunting ranges from around 2.5 percent in the richest countries in the sample to around 54 percent in the poorest countries. The levels of stunting tend to be slightly lower for girls than for boys, as reported in figure C4.3. In the figure, the solid dot indicates the country average, the triangle indicates the average for girls, and the horizontal bar indicates the average for boys. The average stunting rate is 24 percent for boys, compared with 22 percent for girls. Figure C4.4 reports average stunting rates by income group and by World Bank region. Levels tend to be highest in low-income countries, and regional averages are highest in Sub-Saharan Africa and South Asia. 170 A ppendix C : H C I C omp onent Data N otes Table C4.1: Source data for countries with different values in 2018 and back- calculated 2018 2018 VINTAGE 2020 VINTAGE COUNTRY Source Year Value Source Year Value Albania DHS 2009 0.23 DHS 2017 0.11 Burkina Faso SMART 2016 0.27 SMART 2017 0.21 India DHS 2015 0.38 NNS 2017 0.35 Mongolia MICS 2013 0.11 NNS 2016 0.07 Sierra Leone MICS 2017 0.26 MICS 2017 0.31 Tajikistan DHS 2012 0.27 DHS 2017 0.18 Timor-Leste Food and Timor-Leste Demographic Timor-Leste Nutrition Survey, Final 2013 0.50 2016 0.46 and Health Survey 2016 Report 2015 Togo DHS 2014 0.28 MICS 2017 0.24 Figure C4.2: Stunting rates, HCI 2020 0.6 Stunting Rate, circa 2020 0.4 Chad Ethiopia India Indonesia South Africa Ecuador Congo, Rep. 0.2 Senegal Morocco Thailand China Bulgaria 0 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots stunting rates (on the vertical axis) against log GDP per capita at 2011 USD PPP (on the horizontal axis). Stunting rates have seen only modest declines in the last 10 years. Figure C4.5 plots stunting rates in 2020 (on the vertical axis) against rates in 2010 (on the horizontal axis). Of the 42 countries with stunting rates in both 2010 and 2020, roughly 85 percent saw a decrease in stunting (appearing below the dashed 45-degree line), while the remaining T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 171 Figure C4.3: Sex-disaggregated stunting rates 0.6 Sex-disaggregated Stunting Rate, circa 2020 0.4 Chad Ethiopia India Indonesia South Africa Ecuador Congo, Rep. 0.2 Senegal Morocco Thailand China Bulgaria 0 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots sex-disaggregated stunting rates. The solid dot indicates the national average, the triangle is used to show the average value for girls, and the horizontal line shows the average value for boys. Figure C4.4: Stunting rates by income group and region Low income 0.35 Lower-middle income 0.25 High income 0.20 Upper-middle income 0.13 0 .1 .2 .3 .4 Stunting Rate, circa 2020 Sub−Saharan Africa 0.31 South Asia 0.31 East Asia & Pacific 0.24 Middle East & North Africa 0.18 Latin America & Caribbean 0.15 Europe & Central Asia 0.10 0 .1 .2 .3 .4 Stunting Rate, circa 2020 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot regional and income-group average values for stunting. Only two high-income countries have stunting data, Brunei Darussalam and Barbados. 172 A ppendix C : H C I C omp onent Data N otes Figure C4.5: Changes in stunting rates (circa 2010-circa 2020) 0.6 Stunting Rate, circa 2020 0.4 Eswatini Burkina Faso 0.2 Côte d’Ivoire Paraguay North Macedonia 0 0 .2 .4 .6 Stunting Rate, circa 2010 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots stunting rates as used in the calculation of the HCI of 2020 on the vertical axis, and stunting rates as used in the calculation of the HCI of 2010 on the horizontal axis. 15 percent saw increases. The average stunting prevalence in this group of countries dropped from 29 percent in 2010 to 24 percent in 2020. Countries with the largest declines in stunting rates include Eswatini, where rates went down by 14 percentage points (from 40 percent to 26 percent), and India, which experienced a 13-percent- age-point decline (from 48 percent to 35 percent). On the other hand, stunting rates in Angola went up 9 percentage points, from 29 percent to 38 percent. 5. ADULT SURVIVAL RATES The adult survival rate is calculated as the complement of the mortality rate for 15- to 60-year-olds. The mortality rate for 15- to 60-year-olds is the probability of a 15- year- old in a specified year dying before reaching the age of 60 if subject to current age-spe- cific mortality rates. It is frequently expressed as a rate per 1,000 alive at 15, in which case it must be divided by 1,000 to obtain the probability of a 15-year-old dying before age 60. Adult mortality rates are estimated based on prevailing patterns of death rates by age and are reported by the United Nations Population Division (UNPD) for five-year peri- ods. The five-year data are interpolated to arrive at annual estimates to calculate the HCI. The measurement of adult survival rates requires data on death rates by age. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 173 While these are readily available in countries with strong vital registries, such data are missing or incomplete in roughly the poorest quarter of countries. In these countries, the United Nations Population Division estimates death rates by age by linking the lim- ited available age-specific mortality data with model life tables that capture the typical pattern in the distribution of deaths by age. UNPD does not individually report adult mortality rates for countries with less than 90,000 inhabitants. For this reason, data from the UNPD are supplemented with adult mortality rates from the Global Burden of Disease (GBD) project, managed by the Institute of Health Metrics and Evaluation (IHME). Data from this source are used for Dominica and the Republic of the Marshall Islands. Data for Nauru, Palau, San Marino, St. Kitts and Nevis, and Tuvalu come from the World Health Organization (WHO). While there is uncertainty on the primary estimates of mortality as well as the pro- cess for data modeling, uncertainty intervals are not reported in the UNPD data. Here we use uncertainty intervals reported in the GBD modeling process for adult survival rates.38 The point estimates for adult survival rates in these two datasets are quite simi- lar for most countries. The ratio of the upper (lower) bound to the point estimate of the adult survival rate in the GBD data is applied to the point estimate of the adult survival rate in the UNPD and WHO data to obtain upper (lower) bounds. 2020 Update Adult mortality rates for the 2020 update of the HCI come from the 2019 update of the UNPD World Population Prospects estimates, available at the World Population Prospects website, https://population.un.org/wpp/. The GBD data come from the 2017 update—the most recent available—and can be retrieved from the IHME data visu- alization site, http://www.healthdata.org/results/data-visualizations. The WHO data are located on the UN Data platform, https://data.un.org/. Data for five-year periods from the UNPD are interpolated to arrive at annual esti- mates. Data from the GBD and WHO are carried forward up to 10 years to fill gaps in the series. UNPD adult mortality rates for the 2020 HCI come from the most recently available year, as of 2019, while data for the back-calculated 2018 HCI come from 2017. Data for the comparator year of 2010 come from 2010. For countries with data from the GBD, the latest data from 2017 are used to populate the 2020 and back-calculated 2018 rates. For countries with data from the WHO, the most recent estimate to populate the 2020 and back-calculated 2018 rates comes from 2012. Since adult mortality rates are estimated by modeling all available data on adult mortal- ity from vital registration systems, population censuses, household surveys, and sample registration systems combined with model life tables, every new release of data from 38 See Global Burden of Disease (GBD), database, Institute for Health Metrics and Evaluation (IHME), Seattle, http:// www.healthdata.org/gbd. 174 A ppendix C : H C I C omp onent Data N otes Figure C5.1: Comparing original and back-calculated 2018 adult mortality rates 0.5 Central African Republic 0.4 Adult Mortality Rates, 2018 back−calculated Côte d’Ivoire 0.3 Angola Uganda Fiji 0.2 Tonga Kazakhstan Mexico Belarus Brunei Darussalam 0.1 0 0 .1 .2 .3 .4 .5 Adult Mortality Rates, 2018 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots adult rates as used in the HCI of 2018 (on the horizontal axis), and the adult mortality rates used for the back-calculated HCI of 2018 (on the vertical axis). Figure indicates differences that arise due to data updates. the UNPD and GBD updates estimates for all the previous years in the time-series. As a result, data for the same year might differ slightly across updates. Values for adult mortality rates used to produce the back-calculated 2018 HCI are simi- lar to those used in the previous iteration of the HCI, as illustrated in figure C5.1, where data from the two vintages align closely along the 45-degree line for most countries. The figure highlights the 10 countries where adult mortality rates have changed by 30 deaths per 1,000 15-year-olds or more. The largest changes were for Angola (which went from 236 to 279 deaths per 1,000 15-year-olds) and Kazakhstan (which went from 203 to 158 deaths per 1,000 15-year-olds). Figure C5.2 reports the most recent cross-section of adult mortality rates used to calculate the 2020 HCI. Rates range from around 0.039 (39 deaths per 1,000 15-year- olds) in the richest countries to around 0.477 (477 deaths per 1,000 15-year-olds) in the poorest countries. Adult mortality rates tend to be lower for women than for men, as reported in figure C5.3. In the figure, the solid dot indicates the country average, the triangle indicates the average for women, and the horizontal bar indicates the average for men. The average T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 175 Figure C5.2: Adult mortality rates, HCI 2020 0.5 0.4 Chad Adult Mortality Rate, circa 2020 South Africa 0.3 Congo, Rep. Ethiopia 0.2 Senegal India Indonesia Seychelles Bulgaria Ecuador Thailand Hungary 0.1 Oman China Morocco Iran, Islamic Rep.Greece Finland Kuwait Ireland Cyprus Switzerland 0 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots adult mortality rates (on the vertical axis) against log GDP per capita at 2011 USD PPP (on the horizontal axis). Figure C5.3: Sex-disaggregated adult mortality rates 0.6 Sex-disaggregated Adult Mortality Rate, circa 2020 0.4 Chad South Africa Congo, Rep. 0.2 Ethiopia Senegal India Indonesia Seychelles Bulgaria Ecuador Thailand Hungary Oman China Morocco Iran, Islamic Rep.Greece Finland Kuwait Cyprus Ireland Switzerland 0 6 8 10 12 Log GDP per capita at PPP, circa 2020 EAP ECA LAC MENA NA SAR SSA Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figure plots Sex-disaggregated adult mortality rates. The solid dot indicates the national average, the triangle is used to show the average value for girls, and the horizontal line shows the average value for boys. 176 A ppendix C : H C I C omp onent Data N otes Figure C5.4: Adult mortality rates by income group and region Low income 0.25 Lower-middle income 0.20 Upper-middle income 0.14 High income 0.08 0 .05 .1 .15 .2 .25 Adult Mortality Rate, circa 2020 Sub−Saharan Africa 0.26 South Asia 0.16 Latin America & Caribbean 0.14 East Asia & Pacific 0.14 Europe & Central Asia 0.10 Middle East & North Africa 0.09 North America 0.09 0 .05 .1 .15 .2 .25 Adult Mortality Rate, circa 2020 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: The figures plot regional and income-group average values for adult mortality rates. adult mortality rate for men was 0.183 (183 deaths per 1,000 15-year-olds), compared to 0.120 for women. Figure C5.4 reports average adult mortality rates by income group and by World Bank region. Mortality rates tend to be highest in low-income countries, and regional aver- ages are highest in Sub-Saharan Africa and South Asia, reflecting that poor countries continue to bear a disproportionate burden of adult mortality. 6. WORLD BANK-WIDE DATA REVIEW PROCESS AND QUALITY ASSESSMENT The component data of the Human Capital Index 2020 was subject to extensive Bank- wide data review to ensure data timeliness and quality. The review process was conducted between February and July 2020, and was split into two parts. The first part of the data review process (February to May 2020) focused on the enrollment data used to construct estimates of expected years of school and was done by World Bank Program Leaders for Human Development. The second part of the data review process (May to July 2020) focused on the other four index components—child mortality, harmonized test scores, stunting rates, and adult mortality. The enrollment data was validated separately since experience from the first edition of the HCI in 2018 suggested that it required the most intensive review in terms of time and inputs needed from World Bank country teams due to extensive gaps in the data as reported by the UNESCO Institute for Statistics. All component data was reviewed for timeliness and completeness, with gaps filled and revisions made as needed. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 177 7. SUPPLEMENTARY APPENDIX TABLES Table C7.1: Data sources for every level for countries with an absolute change in EYS of at least 0.5 (2018 and 2018 back-calculated) 2018 2018 BACK CALCULATED COUNTRY LEVEL EYS Rate Year Source EYS Rate Year Source Pre-primary 24.9 2016 UIS (ANER) 61.3 2017 UIS (ANER) Primary 99.1 2016 UIS (TNER) 97.9 2017 UIS (TNER) Azerbaijan 11.6 12.4 Lower-secondary 93.8 2016 UIS (TNER) 99.4 2017 UIS (TNER) Upper-secondary 77.4 2017 WB Staff (NER) 77.5 2017 WB Staff (NER) Pre-primary 59.7 2011 UIS (ANER) 41.7 2017 UIS (GER) Primary 92.8 2016 UIS (TNER) 92.5 2017 WB Staff (NER) Bangladesh 11.0 10.2 Lower-secondary 86.8 2016 UIS (TNER) 69.2 2017 UIS (ANER) Upper-secondary 55.4 2016 UIS (TNER) 57.2 2016 UIS (TNER) Pre-primary 95.4 2016 UIS (ANER) 84.0 2017 UIS (ANER) Primary 93.4 2016 UIS (TNER) 88.2 2017 UIS (TNER) Bulgaria 12.9 12.3 Lower-secondary 90.7 2016 UIS (TNER) 87.6 2017 UIS (TNER) Upper-secondary 89.5 2016 UIS (TNER) 90.3 2017 UIS (TNER) Pre-primary 7.9 2012 WB Staff (ANER) 4.0 2013 UIS (NER) Primary 72.1 2014 WB Staff (TNER) 72.1 2014 WB Staff (TNER) Congo, Dem. Rep. 9.2 8.5 Lower-secondary 81.9 2014 WB Staff (TNER) 81.9 2014 WB Staff (TNER) Upper-secondary 74.9 2014 WB Staff (TNER) 53.0 2014 WB Staff (TNER) Pre-primary 21.2 2016 UIS (ANER) 22.2 2017 UIS (ANER) Primary 60.1 2012 WB Staff (TNER) 79.0 2017 UIS (TNER) Côte d’Ivoire 7.0 7.6 Lower-secondary 61.5 2013 WB Staff (TNER) 49.6 2017 UIS (TNER) Upper-secondary 39.0 2013 WB Staff (TNER) 31.7 2017 UIS (TNER) Pre-primary 77.1 2016 UIS (ANER) 87.4 2017 UIS (ANER) Primary 84.9 2016 UIS (TNER) 90.2 2017 UIS (TNER) Dominican Republic 11.3 11.9 Lower-secondary 85.5 2016 UIS (TNER) 87.2 2017 UIS (TNER) Upper-secondary 69.8 2016 UIS (TNER) 71.7 2017 UIS (TNER) Pre-primary 17.0 2011 UIS (ANER) 18.9 2011 UIS (ANER) Primary 64.4 2015 UIS (TNER) 81.4 2017 WB Staff (NER) Eswatini 8.2 6.4 Lower-secondary 75.8 2015 UIS (TNER) 28.3 2017 WB Staff (NER) Upper-secondary 55.9 2015 UIS (TNER) 10.4 2015 WB Staff (NER) Pre-primary 100.0 2015 UIS (GER) 98.8 2017 UIS (ANER) Primary 99.4 2015 UIS (TNER) 99.0 2017 UIS (TNER) Germany 13.9 13.3 Lower-secondary 97.5 2015 UIS (GER) 92.9 2017 UIS (TNER) Upper-secondary 100.0 2015 UIS (GER) 87.6 2017 UIS (TNER) Pre-primary 12.9 2016 UIS (GER) 13.7 2018 WB Staff (ANER) Primary 97.2 2013 UIS (TNER) 88.7 2018 WB Staff (TNER) India 10.2 10.8 Lower-secondary 84.9 2013 UIS (TNER) 62.2 2018 WB Staff (TNER) Upper-secondary 51.0 2013 UIS (TNER) 30.3 2018 WB Staff (TNER) 178 A ppendix C : H C I C omp onent Data N otes Table C7.1: Data sources for every level for countries with an absolute change in EYS of at least 0.5 (2018 and 2018 back-calculated) (continued) 2018 2018 BACK CALCULATED COUNTRY LEVEL EYS Rate Year Source EYS Rate Year Source Pre-primary 36.0 2016 UIS (ANER) 42.4 2016 UIS (ANER) Primary 75.9 2016 UIS (TNER) 88.6 2017 UIS (TNER) Lesotho 8.7 10.0 Lower-secondary 64.3 2016 UIS (TNER) 70.1 2016 UIS (TNER) Upper-secondary 51.3 2016 UIS (TNER) 59.2 2016 UIS (TNER) Pre-primary 26.1 2016 UIS (NER) 60.7 2018 UIS (ANER) Primary 100.0 2016 UIS (GER) 72.5 2018 UIS (TNER) Madagascar 7.5 8.4 Lower-secondary 22.8 2016 UIS (ANER) 63.5 2018 UIS (TNER) Upper-secondary 8.7 2016 UIS (ANER) 31.1 2018 UIS (TNER) Pre-primary 5.5 2008 WB Staff (ANER) 10.5 2015 UIS (GER) Primary 65.5 2016 UIS (TNER) 76.5 2017 UIS (TNER) Mauritania 6.3 7.4 Lower-secondary 45.4 2016 UIS (TNER) 52.3 2017 UIS (TNER) Upper-secondary 28.5 2016 UIS (TNER) 32.9 2017 UIS (TNER) Pre-primary 88.3 2010 UIS (ANER) 55.6 2010 UIS (NER) Primary 90.6 2010 UIS (TNER) 88.7 2010 UIS (TNER) Nicaragua 11.6 10.8 Lower-secondary 82.3 2010 UIS (TNER) 81.8 2010 UIS (TNER) Upper-secondary 63.5 2010 UIS (TNER) 62.4 2010 UIS (TNER) Pre-primary 41.8 2010 UIS (GER) 41.8 2010 UIS (GER) Primary 65.9 2010 UIS (TNER) 66.0 2010 UIS (TNER) Nigeria 8.2 7.3 Lower-secondary 52.5 2013 UIS (GER) 45.0 2016 UIS (GER) Upper-secondary 60.3 2013 UIS (GER) 38.6 2016 UIS (GER) Pre-primary 57.6 2016 UIS (NER) 43.4 2017 WB Staff (ANER) Primary 82.1 2016 UIS (TNER) 77.8 2017 WB Staff (TNER) Pakistan 8.8 9.3 Lower-secondary 53.8 2016 UIS (TNER) 71.4 2017 WB Staff (TNER) Upper-secondary 37.8 2016 UIS (TNER) 55.7 2017 WB Staff (TNER) Pre-primary 78.9 2015 UIS (ANER) 75.6 2017 UIS (ANER) Primary 87.4 2015 UIS (TNER) 84.0 2017 UIS (TNER) Panama 11.3 10.7 Lower-secondary 84.5 2015 UIS (TNER) 84.2 2017 UIS (TNER) Upper-secondary 66.1 2015 UIS (TNER) 55.6 2017 UIS (TNER) Pre-primary 98.6 2008 UIS (GER) 71.4 2016 UIS (ANER) Primary 85.4 2012 UIS (TNER) 84.4 2016 UIS (TNER) Papua New Guinea 8.2 10.3 Lower-secondary 15.6 2012 UIS (NER) 77.1 2016 UIS (TNER) Upper-secondary 22.0 2012 UIS (GER) 50.4 2016 UIS (TNER) Pre-primary 96.5 2016 UIS (ANER) 97.4 2017 UIS (ANER) Primary 99.8 2016 UIS (TNER) 97.2 2017 UIS (TNER) Seychelles 13.7 13.0 Lower-secondary 92.0 2016 UIS (ANER) 91.8 2017 UIS (ANER) Upper-secondary 99.9 2016 UIS (TNER) 83.7 2017 UIS (TNER) T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 179 Table C7.1: Data sources for every level for countries with an absolute change in EYS of at least 0.5 (2018 and 2018 back-calculated) (continued) 2018 2018 BACK CALCULATED COUNTRY LEVEL EYS Rate Year Source EYS Rate Year Source Pre-primary 65.4 2015 UIS (ANER) 55.7 2017 WB Staff (ANER) Primary 75.0 2016 UIS (TNER) 92.9 2017 UIS (TNER) Solomon Islands 9.2 8.7 Lower-secondary 69.0 2007 UIS (TNER) 37.0 2017 WB Staff (NER) Upper-secondary 44.8 2007 UIS (TNER) 28.4 2017 WB Staff (NER) Pre-primary 21.9 2015 UIS (NER) 14.9 2015 UIS (NER) Primary 83.1 2015 UIS (TNER) 92.8 2017 WB Staff (NER) South Africa 9.3 10.2 Lower-secondary 71.1 2015 UIS (TNER) 72.1 2017 UIS (TNER) Upper-secondary 58.4 2015 UIS (TNER) 70.8 2017 WB Staff (NER) Pre-primary 45.0 2014 UIS (ANER) 54.7 2017 UIS (ANER) Primary 91.5 2017 WB Staff (TNER) 80.7 2017 UIS (TNER) Tanzania 7.8 7.2 Lower-secondary 38.7 2017 WB Staff (GER) 27.8 2016 UIS (TNER) Upper-secondary 6.9 2017 WB Staff (GER) 14.2 2016 UIS (TNER) Pre-primary 57.3 2016 UIS (ANER) 43.2 2017 UIS (ANER) Primary 68.9 2016 UIS (TNER) 82.4 2017 UIS (TNER) Timor-Leste 9.9 10.6 Lower-secondary 85.9 2016 UIS (TNER) 85.0 2017 UIS (TNER) Upper-secondary 66.6 2016 UIS (TNER) 74.3 2017 UIS (TNER) Pre-primary 38.5 2014 UIS (GER) 45.8 2015 UIS (GER) Primary 98.7 2014 UIS (TNER) 98.9 2015 UIS (TNER) Tonga 10.9 11.6 Lower-secondary 96.4 2014 UIS (TNER) 95.1 2015 UIS (TNER) Upper-secondary 43.3 2014 UIS (TNER) 62.0 2015 UIS (TNER) Pre-primary 55.9 2015 UIS (NER) 49.7 2015 UIS (NER) Primary 83.6 2015 UIS (TNER) 77.8 2015 UIS (TNER) Vanuatu 10.6 10.1 Lower-secondary 95.3 2015 UIS (TNER) 92.8 2015 UIS (TNER) Upper-secondary 54.9 2015 UIS (TNER) 55.5 2015 UIS (TNER) Pre-primary 89.6 2016 UIS (ANER) 99.7 2017 UIS (ANER) Primary 96.2 2016 WB Staff (NER) 97.9 2017 UIS (TNER) Vietnam 12.3 12.8 Lower-secondary 89.7 2016 WB Staff (NER) 96.7 2017 UIS (NER) Upper-secondary 68.1 2016 WB Staff (NER) 68.1 2016 WB Staff (NER) Pre-primary 64.7 2015 UIS (ANER) 64.4 2017 UIS (ANER) Primary 92.4 2016 UIS (TNER) 97.4 2017 UIS (TNER) West Bank and Gaza 11.4 12.0 Lower-secondary 87.7 2016 UIS (TNER) 92.7 2017 UIS (TNER) Upper-secondary 63.5 2016 UIS (TNER) 68.1 2017 UIS (TNER) Pre-primary 36.4 2013 UIS (ANER) 40.7 2013 UIS (ANER) Primary 87.9 2013 UIS (TNER) 97.6 2013 UIS (TNER) Zimbabwe 10.0 11.1 Lower-secondary 86.9 2013 UIS (TNER) 93.9 2013 UIS (TNER) Upper-secondary 46.7 2013 UIS (TNER) 52.2 2013 UIS (TNER) 180 A ppendix C : H C I C omp onent Data N otes Table C7.2: Data sources for every level of schooling for countries with a decrease in EYS between 2010 and 2020 2010 2020 COUNTRY LEVEL EYS Rate Year Source EYS Rate Year Source Pre-primary 96.9 2010 UIS (ANER) 100.0 2017 UIS (ANER) Primary 96.5 2010 UIS (TNER) 97.0 2017 UIS (TNER) Austria 13.5 13.4 Lower-secondary 98.1 2010 UIS (TNER) 97.5 2017 UIS (TNER) Upper-secondary 93.1 2010 UIS (TNER) 89.4 2017 UIS (TNER) Pre-primary 93.2 2010 UIS (ANER) 84.0 2017 UIS (ANER) Primary 99.2 2010 UIS (TNER) 88.2 2017 UIS (TNER) Bulgaria 12.9 12.3 Lower-secondary 88.6 2010 UIS (TNER) 87.6 2017 UIS (TNER) Upper-secondary 81.1 2010 UIS (TNER) 90.3 2017 UIS (TNER) Pre-primary 99.1 2010 UIS (ANER) 93.7 2017 UIS (ANER) Primary 98.6 2010 UIS (TNER) 98.9 2017 UIS (TNER) Denmark 13.4 13.4 Lower-secondary 99.2 2010 UIS (TNER) 98.4 2017 UIS (TNER) Upper-secondary 85.3 2010 UIS (TNER) 87.9 2017 UIS (TNER) Pre-primary 96.8 2010 UIS (ANER) 98.8 2017 UIS (ANER) Primary 97.2 2010 UIS (TNER) 99.0 2017 UIS (TNER) Germany 13.3 13.3 Lower-secondary 94.5 2010 UIS (TNER) 92.9 2017 UIS (TNER) Upper-secondary 91.4 2010 UIS (TNER) 87.6 2017 UIS (TNER) Pre-primary 94.4 2010 UIS (ANER) 92.7 2017 UIS (ANER) Primary 96.4 2010 UIS (TNER) 98.0 2017 UIS (TNER) Greece 13.4 13.3 Lower-secondary 94.9 2010 UIS (TNER) 92.5 2017 UIS (TNER) Upper-secondary 95.4 2010 UIS (TNER) 92.5 2017 UIS (TNER) Pre-primary 85.5 2010 UIS (ANER) 85.1 2018 UIS (ANER) Primary 84.1 2011 UIS (TNER*) 81.3 2018 UIS (TNER) Guatemala 10.3 9.7 Lower-secondary 78.9 2010 UIS (TNER) 63.8 2018 UIS (TNER) Upper-secondary 38.3 2010 UIS (TNER) 40.6 2018 UIS (TNER) Pre-primary 94.2 2010 UIS (ANER) 87.1 2017 UIS (ANER) Primary 96.5 2010 UIS (TNER) 96.2 2017 UIS (TNER) Hungary 13.0 13.0 Lower-secondary 93.0 2010 UIS (TNER) 94.7 2017 UIS (TNER) Upper-secondary 85.7 2010 UIS (TNER) 86.8 2017 UIS (TNER) T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 181 Table C7.2: Data sources for every level of schooling for countries with a decrease in EYS between 2010 and 2020 (continued) 2010 2020 COUNTRY LEVEL EYS Rate Year Source EYS Rate Year Source Pre-primary 99.6 2010 UIS (ANER) 93.9 2017 UIS (ANER) Primary 99.3 2010 UIS (TNER) 97.2 2017 UIS (TNER) Italy 13.6 13.3 Lower-secondary 95.2 2010 UIS (TNER) 95.9 2017 UIS (TNER) Upper-secondary 93.2 2010 UIS (TNER) 88.9 2017 UIS (TNER) Pre-primary 95.6 2010 WB Staff (ANER) 91.1 2015 WB Staff (ANER) Primary 99.4 2010 WB Staff (TNER) 98.8 2015 WB Staff (TNER) Japan 13.7 13.6 Lower-secondary 99.7 2010 WB Staff (TNER) 99.9 2015 WB Staff (TNER) Upper-secondary 94.7 2010 WB Staff (TNER) 96.4 2015 WB Staff (TNER) Pre-primary 37.9 2010 WB Staff (NER) 36.5 2018 WB Staff (NER) Primary 97.2 2010 WB Staff (NER) 92.4 2018 WB Staff (NER) Jordan 11.8 11.1 Lower-secondary 96.2 2010 WB Staff (NER) 92.4 2018 WB Staff (NER) Upper-secondary 78.8 2010 WB Staff (NER) 70.1 2018 WB Staff (NER) WB Staff Pre-primary 96.9 2013 95.9 2017 UIS (ANER) (ANER**) Primary 99.8 2010 UIS (TNER) 97.6 2017 UIS (TNER) Korea, Rep. 13.7 13.6 Lower-secondary 99.8 2010 UIS (TNER) 94.4 2017 UIS (TNER) Upper-secondary 92.2 2010 UIS (TNER) 99.7 2017 UIS (TNER) Pre-primary 91.7 2012 UIS (ANER*) 81.2 2018 UIS (ANER) Primary 98.3 2010 UIS (TNER) 88.4 2018 UIS (TNER) Kuwait 12.7 12.0 Lower-secondary 92.7 2010 UIS (TNER) 92.1 2015 UIS (TNER) Upper-secondary 71.3 2010 UIS (TNER) 78.1 2015 UIS (TNER) Pre-primary 95.1 2010 UIS (ANER) 98.2 2017 UIS (ANER) Primary 95.7 2010 UIS (TNER) 95.8 2017 UIS (TNER) Luxembourg 12.8 12.4 Lower-secondary 89.4 2010 UIS (TNER) 84.7 2017 UIS (TNER) Upper-secondary 81.8 2010 UIS (TNER) 72.2 2017 UIS (TNER) Pre-primary 92.6 2010 UIS (ANER) 93.3 2018 UIS (ANER) Primary 91.6 2010 UIS (TNER) 91.0 2018 UIS (TNER) Moldova 12.0 11.8 Lower-secondary 87.5 2010 UIS (TNER) 85.0 2018 UIS (TNER) Upper-secondary 66.4 2010 UIS (TNER) 64.5 2018 UIS (TNER) 182 A ppendix C : H C I C omp onent Data N otes Table C7.2: Data sources for every level of schooling for countries with a decrease in EYS between 2010 and 2020 (continued) 2010 2020 COUNTRY LEVEL EYS Rate Year Source EYS Rate Year Source Pre-primary 76.8 2010 UIS (ANER) 75.6 2017 UIS (ANER) Primary 91.4 2010 UIS (TNER) 84.0 2017 UIS (TNER) Panama 11.3 10.7 Lower-secondary 83.3 2010 UIS (TNER) 84.2 2017 UIS (TNER) Upper-secondary 60.5 2010 UIS (TNER) 55.6 2017 UIS (TNER) Pre-primary 77.9 2013 UIS (ANER*) 92.4 2018 UIS (ANER) Primary 97.5 2010 UIS (TNER) 96.8 2018 UIS (TNER) Qatar 12.9 12.8 Lower-secondary 96.6 2011 UIS (TNER*) 89.6 2018 UIS (TNER) Upper-secondary 86.8 2010 UIS (TNER) 83.0 2010 UIS (TNER) WB Staff Pre-primary 78.8 2013 83.9 2018 WB Staff (ANER) (ANER**) Primary 95.8 2010 UIS (TNER) 89.5 2018 WB Staff (TNER) Romania 12.7 11.8 Lower-secondary 91.7 2010 UIS (TNER) 84.9 2018 WB Staff (TNER) Upper-secondary 86.3 2010 WB Staff (TNER) 74.4 2017 WB Staff (TNER) Pre-primary 84.7 2010 UIS (ANER) 82.3 2017 UIS (ANER) Primary 92.0 2010 UIS (TNER) 91.4 2017 UIS (TNER) Slovak Republic 12.7 12.6 Lower-secondary 93.2 2010 UIS (TNER) 93.5 2017 UIS (TNER) Upper-secondary 89.1 2010 UIS (TNER) 89.5 2017 UIS (TNER) Pre-primary 11.1 2015 14.9 2015 UIS (NER) Primary 92.3 2010 WB Staff (NER) 93.1 2018 WB Staff (NER) South Africa 10.2 10.2 Lower-secondary 79.0 2017 UIS (TNER*) 71.2 2017 UIS (TNER) Upper-secondary 69.7 2010 WB Staff (NER) 73.1 2018 WB Staff (NER) WB Staff Pre-primary 67.1 2013 67.6 2017 UIS (ANER) (ANER**) Primary 94.6 2010 UIS (TNER) 92.6 2017 UIS (TNER) Turkey 12.1 12.1 Lower-secondary 96.6 2010 UIS (TNER) 90.4 2017 UIS (TNER) Upper-secondary 74.1 2010 UIS (TNER) 81.5 2017 UIS (TNER) Pre-primary 99.1 2010 UIS (GER) 83.9 2013 UIS (GER) Primary 90.7 2010 UIS (TNER) 91.9 2014 UIS (TNER) Ukraine 13.1 12.9 Lower-secondary 94.8 2010 UIS (TNER) 96.3 2014 UIS (TNER) Upper-secondary 94.4 2010 UIS (TNER) 94.1 2014 UIS (TNER) Notes: *interpolated using the same series; **interpolated using GER; ***extrapolated T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 183 8. HUMAN CAPITAL INDEX AND COMPONENT DATA. Table C8.1: Human Capital Index and Components: HCI 2020, HCI 2018 back-calculated, HCI 2010 HUMAN CAPITAL COMPONENTS OF HUMAN CAPITAL INDEX 2020 INDEX (HCI) Fraction Learning- of Chilren Probability Expected adjusted under HCI 2018 of Survival Years of Harmonized years of Adult sur- 5 not HCI back- HCI Economy to age 5 School Test Scores school vival rate stunted 2020 calculated 2010 Afghanistan 0.94 8.9 355 5.1 0.79 0.62 0.40 0.39 – Albania 0.99 12.9 434 9.0 0.93 0.89 0.63 0.63 0.54 Algeria 0.98 11.8 374 7.1 0.91 0.88 0.53 0.53 0.53 Angola 0.92 8.1 326 4.2 0.73 0.62 0.36 0.36 – Antigua and 0.99 13.0 407 8.4 0.90 – 0.60 0.58 – Barbuda Argentina 0.99 12.9 408 8.4 0.89 0.92 0.60 0.62 0.59 Armenia 0.99 11.3 443 8.0 0.89 0.91 0.58 0.58 – Australia 1.00 13.6 516 11.2 0.95 – 0.77 0.78 0.75 Austria 1.00 13.4 508 10.9 0.94 – 0.75 0.77 0.74 Azerbaijan 0.98 12.4 416 8.3 0.88 0.82 0.58 0.63 0.50 Bahrain 0.99 12.8 452 9.3 0.93 – 0.65 0.66 0.60 Bangladesh 0.97 10.2 368 6.0 0.87 0.69 0.46 0.46 – Belarus 1.00 13.8 488 10.8 0.85 – 0.70 – – Belgium 1.00 13.5 517 11.2 0.93 – 0.76 0.76 0.75 Benin 0.91 9.2 384 5.7 0.77 – 0.40 0.40 0.37 Bhutan 0.97 10.2 387 6.3 0.81 0.79 0.48 – – Bosnia and 0.99 11.7 416 7.8 0.91 0.91 0.58 0.62 – Herzegovina Botswana 0.96 8.1 391 5.1 0.80 – 0.41 0.41 0.37 Brazil 0.99 11.9 413 7.9 0.86 – 0.55 0.55 0.53 Brunei Darussalam 0.99 13.2 438 9.2 0.88 0.80 0.63 – – Bulgaria 0.99 12.3 441 8.7 0.87 0.93 0.61 0.67 0.64 Burkina Faso 0.92 7.0 404 4.5 0.76 0.75 0.38 0.38 0.32 Burundi 0.94 7.6 423 5.2 0.72 0.46 0.39 0.39 0.34 Cambodia 0.97 9.5 452 6.8 0.84 0.68 0.49 0.49 – Cameroon 0.92 8.7 379 5.3 0.70 0.71 0.40 0.39 0.38 Canada 1.00 13.7 534 11.7 0.94 – 0.80 0.80 0.77 Central African 0.88 4.6 369 2.7 0.59 0.59 0.29 – – Republic Chad 0.88 5.3 333 2.8 0.65 0.60 0.30 0.30 0.29 Chile 0.99 13.0 452 9.4 0.92 – 0.65 0.67 0.63 China 0.99 13.1 441 9.3 0.92 0.92 0.65 0.65 – Colombia 0.99 12.9 419 8.6 0.89 0.87 0.60 0.60 0.58 Comoros 0.93 8.2 392 5.1 0.78 0.69 0.40 0.40 – Congo, Dem. Rep. 0.91 9.1 310 4.5 0.75 0.57 0.37 0.36 – Congo, Rep. 0.95 8.9 371 5.3 0.74 0.79 0.42 0.42 0.41 Costa Rica 0.99 13.1 429 9.0 0.92 – 0.63 0.60 0.60 184 A ppendix C : H C I C omp onent Data N otes Table C8.1: Human Capital Index and Components: HCI 2020, HCI 2018 back-calculated, HCI 2010 (continued) HUMAN CAPITAL COMPONENTS OF HUMAN CAPITAL INDEX 2020 INDEX (HCI) Fraction Learning- of Chilren Probability Expected adjusted under HCI 2018 of Survival Years of Harmonized years of Adult sur- 5 not HCI back- HCI Economy to age 5 School Test Scores school vival rate stunted 2020 calculated 2010 Croatia 1.00 13.4 488 10.4 0.92 – 0.71 0.73 0.69 Cyprus 1.00 13.6 502 10.9 0.95 – 0.76 0.75 0.69 Czech Republic 1.00 13.6 512 11.1 0.92 – 0.75 0.76 0.73 Côte d’Ivoire 0.92 8.1 373 4.8 0.66 0.78 0.38 0.37 0.30 Denmark 1.00 13.4 518 11.1 0.93 – 0.76 0.77 0.75 Dominica 0.96 12.4 404 8.0 0.86 – 0.54 0.55 – Dominican 0.97 11.9 345 6.6 0.84 0.93 0.50 0.51 – Republic Ecuador 0.99 12.9 420 8.7 0.88 0.76 0.59 0.60 0.53 Egypt, Arab Rep. 0.98 11.5 356 6.5 0.86 0.78 0.49 0.49 0.48 El Salvador 0.99 10.9 436 7.6 0.82 0.86 0.55 0.54 – Estonia 1.00 13.5 543 11.7 0.90 – 0.78 0.77 0.73 Eswatini 0.95 6.4 440 4.5 0.60 0.74 0.37 0.37 0.31 Ethiopia 0.94 7.8 348 4.3 0.79 0.63 0.38 0.38 – Fiji 0.97 11.3 383 7.0 0.78 0.91 0.51 – – Finland 1.00 13.7 534 11.7 0.93 – 0.80 0.81 0.82 France 1.00 13.8 510 11.3 0.93 – 0.76 0.76 0.76 Gabon 0.96 8.3 456 6.0 0.79 0.83 0.46 0.46 – Gambia, The 0.94 9.5 353 5.4 0.75 0.81 0.42 0.40 0.37 Georgia 0.99 12.9 400 8.3 0.85 – 0.57 0.61 0.54 Germany 1.00 13.3 517 11.0 0.93 – 0.75 0.76 0.76 Ghana 0.95 12.1 307 6.0 0.77 0.82 0.45 0.44 – Greece 1.00 13.3 469 10.0 0.93 – 0.69 0.69 0.71 Grenada 0.98 13.1 395 8.3 0.85 – 0.57 0.54 – Guatemala 0.97 9.7 405 6.3 0.85 0.53 0.46 0.46 0.44 Guinea 0.90 7.0 408 4.6 0.76 0.70 0.37 0.37 – Guyana 0.97 12.2 346 6.8 0.77 0.89 0.50 0.49 – Haiti 0.94 11.4 338 6.1 0.78 0.78 0.45 0.44 – Honduras 0.98 9.6 400 6.1 0.86 0.77 0.48 0.48 – Hong Kong SAR, 0.99 13.5 549 11.9 0.95 – 0.81 0.82 0.78 China Hungary 1.00 13.0 495 10.3 0.88 – 0.68 0.71 0.69 Iceland 1.00 13.5 498 10.7 0.95 – 0.75 0.74 0.76 India 0.96 11.1 399 7.1 0.83 0.65 0.49 0.48 – Indonesia 0.98 12.4 395 7.8 0.85 0.72 0.54 0.54 0.50 Iran, Islamic Rep. 0.99 11.8 432 8.2 0.93 – 0.59 0.59 0.56 Iraq 0.97 6.9 363 4.0 0.84 0.87 0.41 0.40 – Ireland 1.00 13.9 521 11.6 0.94 – 0.79 0.81 0.77 T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 185 Table C8.1: Human Capital Index and Components: HCI 2020, HCI 2018 back-calculated, HCI 2010 (continued) HUMAN CAPITAL COMPONENTS OF HUMAN CAPITAL INDEX 2020 INDEX (HCI) Fraction Learning- of Chilren Probability Expected adjusted under HCI 2018 of Survival Years of Harmonized years of Adult sur- 5 not HCI back- HCI Economy to age 5 School Test Scores school vival rate stunted 2020 calculated 2010 Israel 1.00 13.8 481 10.6 0.95 – 0.73 0.76 0.72 Italy 1.00 13.3 493 10.5 0.95 – 0.73 0.75 0.75 Jamaica 0.99 11.4 387 7.1 0.86 0.94 0.53 0.54 – Japan 1.00 13.6 538 11.7 0.95 – 0.80 0.84 0.82 Jordan 0.98 11.1 430 7.7 0.89 – 0.55 0.55 0.56 Kazakhstan 0.99 13.7 416 9.1 0.84 0.92 0.63 0.78 0.59 Kenya 0.96 11.6 455 8.5 0.77 0.74 0.55 0.54 – Kiribati 0.95 11.2 411 7.4 0.81 – 0.49 0.47 – Korea, Rep. 1.00 13.6 537 11.7 0.94 – 0.80 0.83 0.82 Kosovo 0.99 13.2 374 7.9 0.91 – 0.57 0.57 – Kuwait 0.99 12.0 383 7.4 0.94 – 0.56 0.56 0.57 Kyrgyz Republic 0.98 12.9 420 8.7 0.85 0.88 0.60 0.59 – Lao PDR 0.95 10.6 368 6.3 0.82 0.67 0.46 0.46 – Latvia 1.00 13.6 504 11.0 0.84 – 0.71 0.74 0.68 Lebanon 0.99 10.2 390 6.3 0.93 – 0.52 0.52 – Lesotho 0.92 10.0 393 6.3 0.52 0.65 0.40 0.40 0.34 Liberia 0.93 4.2 332 2.2 0.78 0.70 0.32 0.32 – Lithuania 1.00 13.8 496 11.0 0.84 – 0.71 0.73 0.69 Luxembourg 1.00 12.4 493 9.8 0.94 – 0.69 0.69 0.70 Macao SAR, China 0.99 12.9 561 11.6 0.96 – 0.80 0.76 0.65 Madagascar 0.95 8.4 351 4.7 0.80 0.58 0.39 0.39 0.39 Malawi 0.95 9.6 359 5.5 0.74 0.61 0.41 0.41 0.36 Malaysia 0.99 12.5 446 8.9 0.88 0.79 0.61 0.63 0.58 Mali 0.90 5.2 307 2.6 0.75 0.73 0.32 0.32 – Malta 0.99 13.4 474 10.2 0.95 – 0.71 0.71 0.68 Marshall Islands, Rep. 0.97 9.4 375 5.7 0.70 0.65 0.42 0.40 – Mauritania 0.92 7.7 342 4.2 0.80 0.77 0.38 0.37 – Mauritius 0.98 12.4 473 9.4 0.86 – 0.62 0.62 0.60 Mexico 0.99 12.8 430 8.8 0.86 0.90 0.61 0.61 0.59 Micronesia, Fed. Sts. 0.97 11.8 380 7.2 0.84 – 0.51 0.47 – Moldova 0.98 11.8 439 8.3 0.84 0.94 0.58 0.58 0.56 Mongolia 0.98 13.2 435 9.2 0.80 0.91 0.61 0.62 – Montenegro 1.00 12.8 436 8.9 0.91 0.91 0.63 0.62 0.59 Morocco 0.98 10.4 380 6.3 0.93 0.85 0.50 0.49 0.47 Mozambique 0.93 7.6 368 4.5 0.68 0.58 0.36 0.36 – Myanmar 0.95 10.0 425 6.8 0.80 0.71 0.48 0.47 – Namibia 0.96 9.4 407 6.1 0.71 0.77 0.45 0.45 0.39 186 A ppendix C : H C I C omp onent Data N otes Table C8.1: Human Capital Index and Components: HCI 2020, HCI 2018 back-calculated, HCI 2010 (continued) HUMAN CAPITAL COMPONENTS OF HUMAN CAPITAL INDEX 2020 INDEX (HCI) Fraction Learning- of Chilren Probability Expected adjusted under HCI 2018 of Survival Years of Harmonized years of Adult sur- 5 not HCI back- HCI Economy to age 5 School Test Scores school vival rate stunted 2020 calculated 2010 Nauru 0.97 11.7 347 6.5 0.93 – 0.51 – – Nepal 0.97 12.3 369 7.2 0.86 0.64 0.50 0.50 – Netherlands 1.00 13.9 520 11.5 0.95 – 0.79 0.80 0.80 New Zealand 0.99 13.7 520 11.4 0.94 – 0.78 0.77 0.78 Nicaragua 0.98 10.8 392 6.7 0.85 0.83 0.51 0.51 – Niger 0.92 5.5 305 2.7 0.77 0.52 0.32 0.32 – Nigeria 0.88 10.2 309 5.0 0.66 0.63 0.36 0.35 – North Macedonia 0.99 11.0 414 7.3 0.91 0.95 0.56 0.54 0.54 Norway 1.00 13.7 514 11.2 0.94 – 0.77 0.77 0.77 Oman 0.99 12.8 424 8.6 0.91 – 0.61 0.61 0.55 Pakistan 0.93 9.4 339 5.1 0.85 0.62 0.41 0.40 – Palau 0.98 11.7 463 8.7 0.87 – 0.59 0.57 – Panama 0.98 10.7 377 6.5 0.89 – 0.50 0.51 0.51 Papua New Guinea 0.95 10.3 363 6.0 0.78 0.51 0.43 0.42 – Paraguay 0.98 11.3 386 7.0 0.86 0.94 0.53 0.53 0.51 Peru 0.99 13.0 415 8.6 0.89 0.88 0.61 0.59 0.55 Philippines 0.97 12.9 362 7.5 0.82 0.70 0.52 0.55 – Poland 1.00 13.4 530 11.4 0.89 – 0.75 0.76 0.70 Portugal 1.00 13.9 509 11.3 0.93 – 0.77 0.78 0.74 Qatar 0.99 12.8 427 8.8 0.96 – 0.64 0.63 0.59 Romania 0.99 11.8 442 8.4 0.88 – 0.58 0.59 0.60 Russian Federation 0.99 13.7 498 10.9 0.80 – 0.68 0.73 0.60 Rwanda 0.96 6.9 358 3.9 0.81 0.62 0.38 0.38 – Samoa 0.98 12.2 370 7.2 0.89 0.95 0.55 0.52 – Saudi Arabia 0.99 12.4 399 7.9 0.92 – 0.58 0.58 0.55 Senegal 0.96 7.3 412 4.8 0.83 0.81 0.42 0.42 0.39 Serbia 0.99 13.3 457 9.8 0.89 0.94 0.68 0.76 0.65 Seychelles 0.99 13.1 463 9.7 0.85 – 0.63 0.63 0.57 Sierra Leone 0.89 9.6 316 4.9 0.63 0.71 0.36 0.35 – Singapore 1.00 13.9 575 12.8 0.95 – 0.88 0.89 0.85 Slovak Republic 0.99 12.6 485 9.8 0.90 – 0.66 0.68 0.68 Slovenia 1.00 13.6 521 11.4 0.93 – 0.77 0.79 0.75 Solomon Islands 0.98 8.3 351 4.7 0.86 0.68 0.42 0.43 – South Africa 0.97 10.2 343 5.6 0.69 0.73 0.43 0.42 0.43 South Sudan 0.90 4.7 336 2.5 0.68 0.69 0.31 0.31 – Spain 1.00 13.0 507 10.5 0.95 – 0.73 0.74 0.71 Sri Lanka 0.99 13.2 400 8.5 0.90 0.83 0.60 0.59 – T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 187 Table C8.1: Human Capital Index and Components: HCI 2020, HCI 2018 back-calculated, HCI 2010 (continued) HUMAN CAPITAL COMPONENTS OF HUMAN CAPITAL INDEX 2020 INDEX (HCI) Fraction Learning- of Chilren Probability Expected adjusted under HCI 2018 of Survival Years of Harmonized years of Adult sur- 5 not HCI back- HCI Economy to age 5 School Test Scores school vival rate stunted 2020 calculated 2010 St. Kitts and Nevis 0.99 13.0 409 8.5 0.88 – 0.59 0.57 – St. Lucia 0.98 12.7 418 8.5 0.87 0.98 0.60 0.59 – St. Vincent and the 0.98 12.3 391 7.7 0.83 – 0.53 0.54 – Grenadines Sudan 0.94 7.1 380 4.3 0.79 0.62 0.38 0.38 – Sweden 1.00 13.9 519 11.6 0.95 – 0.80 0.80 0.76 Switzerland 1.00 13.3 515 10.9 0.95 – 0.76 0.77 0.77 Tajikistan 0.97 10.9 391 6.8 0.87 0.82 0.50 0.54 – Tanzania 0.95 7.2 388 4.5 0.78 0.68 0.39 0.39 – Thailand 0.99 12.7 427 8.7 0.87 0.89 0.61 0.62 0.58 Timor-Leste 0.95 10.6 371 6.3 0.86 0.54 0.45 0.45 0.41 Togo 0.93 9.7 384 6.0 0.74 0.76 0.43 0.42 0.37 Tonga 0.98 11.6 386 7.1 0.83 0.92 0.53 0.52 – Trinidad and Tobago 0.98 12.4 458 9.1 0.85 – 0.60 0.60 0.55 Tunisia 0.98 10.6 384 6.5 0.91 0.92 0.52 0.51 0.53 Turkey 0.99 12.1 478 9.2 0.91 0.94 0.65 0.63 0.63 Tuvalu 0.98 10.8 346 6.0 0.79 – 0.45 0.44 – Uganda 0.95 6.8 397 4.3 0.74 0.71 0.38 0.38 0.34 Ukraine 0.99 12.9 478 9.9 0.81 – 0.63 0.64 0.63 United Arab 0.99 13.5 448 9.6 0.94 – 0.67 0.68 0.62 Emirates United Kingdom 1.00 13.9 520 11.5 0.93 – 0.78 0.78 0.77 United States 0.99 12.9 512 10.6 0.89 – 0.70 0.71 0.69 Uruguay 0.99 12.2 438 8.6 0.89 – 0.60 0.60 0.59 Uzbekistan 0.98 12.0 474 9.1 0.87 0.89 0.62 – – Vanuatu 0.97 10.1 348 5.6 0.87 0.71 0.45 0.44 – Vietnam 0.98 12.9 519 10.7 0.87 0.76 0.69 0.69 0.66 West Bank and 0.98 12.2 412 8.0 0.89 0.93 0.58 0.57 – Gaza Yemen, Rep. 0.95 8.1 321 4.2 0.80 0.54 0.37 0.37 – Zambia 0.94 8.8 358 5.0 0.73 0.65 0.40 0.39 – Zimbabwe 0.95 11.1 396 7.0 0.65 0.77 0.47 0.46 0.41 Source: World Bank calculations based on the 2020 update of the Human Capital Index. Notes: This table reports the components and overall index scores for the Human Capital Index 2020, the back-calculated HCI 2018 and the HCI 2010. The Human Capital Index ranges between 0 and 1. The index is measured in terms of the productivity of the next generation of workers relative to the benchmark of complete education and full health. An economy in which a child born today can expect to achieve complete education and full health will score a value of 1 on the index. Empty cells indicate missing data. 188 A ppendix C : H C I C omp onent Data N otes 9. REFERENCES Altinok, N., N. Angrist, and H. A. Patrinos. 2018. “Global Data Set on Education Quality (1965–2015).” Policy Research Working Paper 8314, World Bank, Washington, DC. Angrist, N., S., Djankov, P. K, Goldberg, and H. A. Patrinos. 2019. “Measuring Human Capital.” World Bank Policy Research Working Paper No. 8742, World Bank, Washington, DC. Atamanov, A., Castaneda Aguilar, R., Diaz-Bonilla, C., Jolliffe, D., Lakner, C., Mahler, D., Montes, J., Moreno Herrera, L., Newhouse, D., Nguyen, M., Prydz, E., Sangraula, P., Tandon, S., Yang, J. 2019. “September 2019 PovcalNet Update: What’s New (English).” Global Poverty Monitoring Technical Note, no. 10, World Bank, Washington, DC Case, A., and C.H. Paxson. 2008. “Stature and Status: Height, Ability, and Labour Market Outcomes.” Journal of Political Economy 116 (3): 499–532. Caselli, F. 2005. “Accounting for Cross-Country Income Differences.” In Handbook of Economic Growth, vol. 1A, edited by Philippe Aghion and Steven N. Darlauf, 679– 741. Amsterdam: Elsevier. Collin, M., and D. Weil. 2018. “The Effect of Increasing Human Capital Investment on Economic Growth and Poverty: A Simulation Exercise.” World Bank Policy Research Working Paper No. 8590, World Bank, Washington, DC D’Souza, R., R. Gatti, and A. Kraay. 2019. “A Socioeconomic Disaggregation of the World Bank Human Capital Index.” Policy Research Working Paper 9020, World Bank, Washington, DC. Filmer, D., H. Rogers, N. Angrist, S. Sabarwal, and R. Gatti. 2018. “Learning-Adjusted Years of Schooling (LAYS): Defining a New Macro Measure of Education.” World Bank Policy Research Working Paper No. 8591, World Bank, Washington, DC Galasso, E., and A. Wagstaff. 2016. “The Economic Costs of Stunting and How to Reduce Them.” With Sophie Naudeau and Meera Shekar. Policy Research Note 5 (March), World Bank, Washington, DC. Gao, Q., Y. Shi, H. Yi, C. Abbey, and S. Rozelle. 2017. “Reading Achievement in China’s Rural Primary Schools: A Study of Three Provinces.” Stanford University, Freeman Spogli Institute Working Paper. Global Burden of Disease (GBD), database, Institute for Health Metrics and Evaluation (IHME), Seattle, http://www.healthdata.org/gbd.Horton, S., and R. Steckel. 2011. Global Economic Losses Attributable to Malnutrition 1900–2000 and Projections to 2050. Assessment Paper: Malnutrition, Copenhagen Consensus on Human Challenges, Tewksbury, MA. Horton, S., and R. Steckel. 2011. “Global Economic Losses Attributable to Malnutrition 1900–2000 and Projections to 2050.” Assessment Paper: Malnutrition, Copenhagen Consensus on Human Challenges, Tewksbury, MA. Hsieh, C. T., and P. Klenow. 2010. “Development Accounting.” American Economic Journal: Macroeconomics 2 (1): 207–23. T H E H U M AN CAP ITAL IN DEX 2 0 2 0 UPDAT E: HUM A N CA PI TA L I N T HE T I M E O F C OV I D- 1 9 189 Joint Child Malnutrition Estimates ( JME): Levels and trends (2020 edition), data- base, https://data.unicef.org/resources/jme-report-2020.Jorgenson, D., and B. M. Fraumeni. 1989. “The Accumulation of Human and Nonhuman Capital, 1948– 1984.” In: The Measurement of Saving, Investment, and Wealth, edited by Robert E. Lipsey and Helen Stone Tice, 227–86. Studies in Income and Wealth Series, vol. 52. Cambridge. MA: National Bureau of Economic Research; Chicago: University of Chicago Press. Kraay, A. 2018. “Methodology for a World Bank Human Capital Index.” World Bank Policy Research Working Paper No. 8593, World Bank, Washington, DC. Marini, A., and C. Rokx. 2017. “Standing Tall: Peru’s Success in Overcoming Its Stunting Crisis.” YouTube video, December 4. With P. Gallagher, World Bank, Washington, DC. Mincer, J. 1958. “Investment in Human Capital and Personal Income Distribution.” Journal of Political Economy 66 (4): 281–302. Montenegro, C. E., and H. A. Patrinos. 2014. “Comparable Estimates of Returns to Schooling around the World.” Policy Research Working Paper 7020, World Bank, Washington, DC. Patrinos, H. A., and N. Angrist. 2018. “A Global Dataset on Education Quality: A Review and an Update (1965–2018).” World Bank Policy Research Working Paper No. 8592, World Bank, Washington, DC. UNECE (United Nations Economic Commission for Europe). 2016. Guide on Measuring Human Capital. Geneva: United Nations. https://www.unece.org/index. php?id=44704. UNICEF (United Nations Children’s Fund). 2013. A Passport to Protection: A Guide to Birth Registration Programming. New York: UNICEF. UNICEF (United Nations Children’s Fund), World Health Organization, and World Bank. 2012. UNICEF-WHO-World Bank Joint Child Malnutrition Estimates (Global Database on Child Growth and Malnutrition). New York, UNICEF; Geneva, WHO; Washington, DC, World Bank. United Nations Interagency Group for Child Mortality Estimation (UNIGME, 2019). “Levels and Trends in Child Mortality, Report 2019,” United Nations Children’s Fund, New York. United Nations Statistics Division, December 2014. Coverage of Birth and Death Registration (database). http://unstats.un.org/unsd/demographic/CRVS/CR_cov- erage.htm. UNU-IHDP (United Nations University–International Human Dimensions Programme on Global Environmental Change) and UNEP (United Nations Environment Programme. 2012. Inclusive Wealth Report 2012: Measuring Progress toward Sustainability. Cambridge, UK: Cambridge University Press. Weil, D. 2007. “Accounting for the Effect of Health on Economic Growth.” Quarterly Journal of Economics 122 (3): 1265–1306. 190 A ppendix C : H C I C omp onent Data N otes World Bank. 2018a. World Development Report 2018: Learning to Realize Education’s Promise. Washington, DC: World Bank. World Bank. 2018b. The Human Capital Project. World Bank, Washington, DC. World Health Organization. 2009. The WHO Multicentre Growth Reference Study (MGRS). Geneva, World Health Organization. World Population Prospects 2019, database, https://population.un.org/wpp/ Yang, H., de Onis, M. 2008. “Algorithms for converting estimates of child malnutrition based on the NCHS reference into estimates based on the WHO Child Growth Standards.” BMC Pediatr, 8 (19).