Source: World Bank calculations based on the 020 update of the Human Capital Index (hci) for hci data and the World Development Indicators and Penn World Tables for per capita gdp data

Source: World Bank calculations based on the 2020 update of the Human Capital Index (HCI) for HCI data and the World Development Indicators and Penn World Tables 9.1 for per capita GDP data. Note: The figure plots country-level HCI on the y-axis and GDP per capita in PPP on the x-axis, in constant 2011 dollars, for most recently available data as of 2019. Per capita GDP data for South Sudan are not available. 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 better (worse) in the HCI than their level of GDP would predict. Economies above the 95th and below the 5th percentile in distance to the regression fitted line are labeled. HCI = Human Capital Index; PPP = purchasing power parit

Source: World Bank calculations based on the 2020 update of the Human Capital Index (HCI). Poverty values come from Corral et al. (2020) and are calculated before COVID-19. Note: The horizontal axis represents the share of the global population accounted for by the countries sorted by their HCI value. HCI = Human Capital Index; PPP = purchasing power parity

Two elements help explain how different dimensions of human capital contribute to differences in the HCI scores. The first are the weights of the health and education components of the HCI, reflecting the empirical literature on the contribution of health and education to earnings (box 1.2 and appendix A). Second, the components have different distributions, globally and by country income groupings, according to the World Bank’s most recent classification. For example, the variation of child survival is nine times larger among low-income than among high-income economies, where child survival is uniformly close to 100 percent (figure 1.3). A simple decomposition exercise can help account for differences in the HCI across country income groups.4 Consider the HCI difference between the typical low-income and high-income economy, which is about 0.33 (figure 1.4). Of these 33 HCI points, almost 25 are accounted for by the differences in expected years of school (EYS) and harmonized test scores. Overall, differences in the quality and quantity of schooling account for the largest share of index differences across country income groups, ranging from 65 to 85 percent. There is also considerable heterogeneity within country income groups, and the difference in HCI between the economy with the lowest HCI and the economy with the highest HCI in each income group rivals the difference between income groups and, in some cases, exceeds it. For example, the difference in the HCI between the top and bottom performers among high-income economies is roughly 0.38, or 38 HCI points, which compares with a difference of 33 points between the average HCI values of high- and low-income economies. Overall, both within and across all groups, education still accounts for the largest share of the differences observed between top and bottom performers (figure 1.5); however, education accounts for a smaller share as one moves down income groups, falling from roughly 90 percent among high-income to 60 percent among low-income economies. In contrast, differences in child survival rates account for less of the difference in HCI scores among highincome economies, largely because economies in this group are close to universal child survival. The same is true for the health component, with stunting and adult survival taken together for easy comparison. Health differences explain a lower share of HCI differences as one moves from low- to high-income economies, because health outcomes tend to be uniformly better as economies get richer.5 These results reflect the fact that, within the high-income group, values for health and survival components in most economies are close to the frontier, whereas there is still considerable variation in test scores (see the box plots in figure 1.3). Gaps in human capital outcomes between rich and poor people within economies can be quite large. A socioeconomic disaggregation of the HCI, constructed using comparable survey data for 50 low- and middle-income economies, reveals that differences across socioeconomic quintiles within economies account for nearly one-third of the total variation in human capital (D’Souza, Gatti, and Kraay 2019). Outcomes can also vary across rural-urban status, as in the case of Romania. Some of that country’s counties have urban areas with learning outcomes as high as top performers in Europe, whereas some rural areas rank at par with economies in the bottom third of the HCI distribution (World Bank 2020a). Some of these within-country differences align with ethnic divides. For example, in Vietnam, survey data from 2014 disaggregated by ethnic group show that ethnic minorities have an HCI score of 0.62, compared with a score of 0.75 for the ethnic majority. At 32 percent, stunting rates are two times larger among ethnic minorities than among the majority. School enrollment also lags among ethnic minorities relative to their majority peers by 30 percentage points (World Bank 2019b)

Box 1.3: Limitations of the Human Capital Index

Like all cross-country benchmarking exercises, the Human Capital Index (HCI) has limitations. Components of the HCI such as stunting and test scores are measured only infrequently in some economies and not at all in others. Data on test scores come from different international testing programs and 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 education system in an economy, to the extent that test-takers are not representative of the population of all students. Reliable measures of the quality of tertiary education that are comparable across most economies 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 estimate expected years of school often have many gaps and are reported with significant lags. Socioemotional skills are not explicitly captured. In terms of health, child and adult survival rates are imprecisely estimated in economies where vital registries are incomplete or nonexistent. These limitations have implications not only for the construction of the 2020 update but also for 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 it is not a finely graduated measurement that can distinguish small differences between economies. 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 economies. Although the HCI combines education and health into a single measure, it is too blunt a tool to inform the cost-effectiveness of policy interventions in these areas, which should instead be assessed through 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 economies, it does not capture cross-country differences in how well economies 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.

1.3.1 HCI components and data sources The components of the HCI are built using publicly available official data, primarily from administrative sources. The data are subject to a careful vetting process with World Bank country teams and, at the discretion of country teams, with line ministry counterparts. These data and the relevant definitions are described in the text that follows and in more in detail in appendix C.

Child survival 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 that a child born in a specified year will die 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) using mortality as recorded in household surveys and vital registries. For the 2020 update of the HCI, under-5 mortality rates come from the September 2019 update of the IGME estimates and are available on the IGME website.


Expected years of school The EYS component of the HCI captures the number of years of school a child born today can expect to obtain by age 18, given the prevailing pattern of enrollment rates in her economy. Conceptually, EYS is the sum of enrollment rates by age from ages 4 to 17. Because age-specific enrollment rates are neither broadly nor systematically available, data on enrollment rates by level of school are used to approximate enrollment rates in different age brackets. Preprimary enrollment rates approximate the enrollment rates for 4- and 5-year-olds, primary enrollment rates approximate the rates 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. Crosscountry definitions in school starting ages and the duration of the various levels of school imply that these rates will only be approximations of the number of years of school a child can expect to complete by age 18. Enrollment rates for 2020 for each school level and for different enrollment rate types are obtained from the United Nations Educational, Scientific and Cultural Organization’s Institute for Statistics.7 These data are then complemented with inputs from World Bank teams working on specific countries to validate the data and provide more recent values when available.8 Harmonized test scores The school quality indicator is based on a largescale effort to harmonize international student achievement tests from several multicountry testing programs to produce 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. The dataset harmonizes scores from three major international testing programs: the Trends in International Mathematics and Science Study (TIMSS), the Progress in International Reading Literacy Study (PIRLS), and the Programme for International Student Assessment (PISA). It further includes four 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 2020 update of the Global Dataset on Education Quality extends the database to 184 economies from 2000 to 2019, drawing on a large-scale effort by the World Bank to collect global learning data. Updates to the database come from new data from PISA 2018, PISA for Development (PISA-D),9 PILNA, and EGRA. The database adds 20 new economies,10 bringing 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 economies.11 Since the launch of the HCI in 2018, a complementary measure has been created to address foundational skills and to help economies prioritize their response to HCI and learning-adjusted years of schooling (LAYS) scores: Learning Poverty represents the share of 10-year-olds who cannot read and understand a simple text (see box 1.4). The correlation between Learning Poverty and LAYS is high, in the range of −0.90. The Learning Poverty measure is available for 113 of the economies in the HCI 2020. Fraction of children under 5 not stunted 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 median for their ages. The median and standard deviations are set by the World Health Organization (WHO) for normal healthy child development (World Health Organization 2009). 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, in addition to the adult survival rate, as a proxy for latent health of the population in economies where stunting data are available, as discussed in the next section. Stunting rates for this edition of the HCI come from the March 2020 update of the


Source: World Bank calculations based on the 2020 update of the Human Capital Index (HCI). Note: 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.