Policy Research Working Paper 9458 Measuring Human Capital in Europe and Central Asia Asli Demirgüç-Kunt Iván Torre Europe and Central Asia Region Office of the Chief Economist October 2020 Policy Research Working Paper 9458 Abstract This paper outlines an extension of the Human Capital (ECA-HCI) extends the Human Capital Index by adding Index that addresses the specific challenges in education a measure of quality-adjusted years of higher education and health faced by countries in Europe and Central Asia. to the original education component, and it includes the Good basic education will not be enough, as job markets prevalence of three adult health risk factors—obesity, smok- today demand higher levels of human capital than in ing, and heavy drinking—as an additional proxy for latent the past. As the region’s population becomes older, it is health status. This extension of the Human Capital Index important that adults remain healthy to ensure productive could also be useful for assessing the state of human capital aging. The Europe and Central Asia Human Capital Index in middle-income countries in general. This paper is a product of the Office of the Chief Economist, Europe and Central Asia Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at ademirguckunt@worldbank.org and itorre@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Measuring Human Capital in Europe and Central Asia Asli Demirgüç-Kunt and Iván Torre Keywords: human capital, education, health, Europe and Central Asia. JEL: I1, I2, O1, O4 *The authors are at the World Bank. This paper’s findings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. We thank Aart Kraay for his guidance and advice, and Tania Dmytraczenko, Roberta Gatti, Harry Patrinos, Fadia Saadah, Gil Shapira and Christel Vermeersch for useful comments. Sharanya Venu Pillai provided excellent assistance. 1. Introduction In 2018, the World Bank launched the Human Capital Project (HCP), an initiative aimed at raising awareness among policy makers about the importance of investing in human capital. The HCP includes an advocacy component which features the Human Capital Index (HCI), a measure of the human capital that a child born today can expect to attain by age 18, given the risks of poor health and poor education that prevail in the country where she lives (Kraay, 2019). The HCI quantifies the trajectory from birth to adulthood in terms of the consequences for productivity by means of three components: (1) a measure of whether children survive from birth to school age (age 5); (2) a measure of expected years of basic education (primary and secondary), adjusted for quality; and (3) two broad measures of health: child stunting rates and adult survival from age 15 to age 60. The index is constructed so that a value of 1 represents the productivity in adulthood of a child born today if he or she enjoyed complete education and full health until age 18. Countries are measured with respect to this benchmark; the value of the index can thus be interpreted as a percentage of that productivity level. While useful in a global context, the original version of the HCI may not adequately reflect the education and health challenges that are relevant for specific regions of the world. Countries in Europe and Central Asia provide their citizens relatively good basic education and health services; the region’s citizens begin their productive life in a much better position than their peers in other regions of the world. But job markets today demand higher levels of human capital than in the past. Good basic education will not be enough; higher education institutions must prepare students for the challenges the future of work may hold. Health care systems will need to ensure that citizens remain healthy throughout their adult life, as, more and more, learning and skill acquisition will take place along an individual’s life cycle, not just in the initial years of life. This is also more important as the region’s workforce is becoming, on average, older, and therefore improving adult health will be needed to ensure a productive aging for the region’s population. This paper outlines an extension of the HCI which addresses the relevant education and health challenges of Europe and Central Asia, namely by including higher education in the education component of the index and by looking at three crucial adult health risk factors -obesity, heavy alcohol consumption and tobacco smoking- in the health component. This extension could also be useful for assessing the state of human capital in middle income countries in general, particularly for those where basic education attainment and child health are less of a concern but where significant challenges remain as young people transition into the labor market. 2 This paper is organized as follows. Section 2 presents the main analytical framework of the ECA-HCI, the proposed extension of the original HCI. Section 3 discusses the education component and section 4 discusses the health component. Section 5 presents the overall results and section 6 concludes. 2. Main Framework The basic structure of the Human Capital Index (HCI) is made up of three components: ∗ ∗) = × ( − ) × ( − (1) ∗ The first term captures forgone productivity caused by child mortality. The second term captures forgone productivity as a result lack of full education, where SNG refers to the schooling level of the generation of children born today and S* refers to the full education benchmark. The productivity return to education is measured by parameter φ. The third terms captures forgone productivity as a result of lack of proper health, where ZNG refers to the expected adult health status of the generation of children born today and Z* refers to the full health benchmark. The productivity return to good health is measured by parameter γ. The HCI’s measure of child mortality is the probability of survival to age five. The education component of the HCI uses learning-adjusted years of schooling, a quality-adjusted measure of years of basic education. The benchmark is set at 14 years of schooling, equivalent to the whole cycle of primary and secondary education plus two years of preprimary education. The parameter φ is set at 0.08, based on estimations of the average return of one year of basic education. The health component of the HCI uses child stunting (when available) and the adult survival rate (the probability that a child age 15 reaches age 60) as health status indicators. The benchmark is zero stunting and 100 percent adult survival rate. To establish a quantifiable productivity return to good health, both variables are transformed into implied adult height in centimeters, which has a productivity return of 0.034 per centimeter. Adult height is implied to be the most relevant proxy variable for latent health status (captured by Z in the equation above). The value of γ is 0.35 for child stunting and 0.65 for the adult survival rate. The HCI is calculated using the following formula: 1− 5 ∗ = × 0.08(−14 ) × (0.35( −1)+0.65(−1))/2 (2) 1 This paper outlines an alternative specification that may be particularly relevant for the education and health challenges faced in Europe and Central Asia. For the education component, we add higher education in addition to basic education. For the health component, we use a proxy of latent adult health status (based on the incidence of obesity, smoking, and alcoholism), along with the outcome proxy based on child 3 stunting and adult survival rate used in the original HCI. The basic formulation of the Europe and Central Asia HCI (ECA-HCI) is as follows: (−∗ )+ (−∗ ) (−∗ )+(− ∗ ) − = × × 2 (3) ∗ where B refers to the quality-adjusted basic education schooling level of the generation of children born today, with an associated productivity return captured by parameter η and full basic education benchmark B*; C refers to the quality-adjusted higher education schooling level, with an associated productivity return captured by parameter ω and full higher education benchmark C*; RF refers to the prevalence of adult health risk factors (namely the share of non-obese individuals in the adult population, the share of adult nonsmokers, and the share of adults who report no heavy drinking), with an associated productivity return captured by parameter γRF. The benchmarks for these shares are set to 100 percent non-obese, nonsmokers, and non–heavy drinkers. O refers to the value of the relevant health outcomes (adult survival rate and child stunting); γo refers to their productivity effects, estimated via their relationship with adult height, as in the original HCI. 3. Education Component The 2019 World Development Report highlights the changing nature of work across the globe. In high- income countries, which include most of the countries in Europe and Central Asia, having a good basic education will not be enough for individuals to be productively included in the labor market in the next decades; higher education of good quality will be necessary for the next generations to be productive workers. The education component of the ECA-HCI therefore extends the original education component by adding a measure of quality-adjusted years of higher education (QAYH) to the measure of learning- adjusted years of basic education. Like learning-adjusted years of basic education (LAYS), QAYH measures both quantity and quality. The basic formulation of the education component of the ECA-HCI is the following: ∗ )+(− ∗ ) − = (− (4) Where η and ω are the productivity returns of one additional year of quality basic and higher education respectively, and LAYS* and QAYH* are the benchmark number of years equivalent to full basic and higher education respectively. As shown in equation 4, the education component of the ECA-HCI includes two subcomponents. The first measures the basic education schooling level expected for the generation of children born today. This 4 component is the same as the overall education component in the standard version of the HCI. The main variable is learning-adjusted years of education, a quality-adjusted measure of schooling years in basic education. The benchmark (LAYS) is set at 14 years of basic education. The associated return in productivity terms (η) is set at 0.08. The second component focuses on higher education. A quality-adjusted measure of years of higher education requires two inputs: a measure of expected years of higher education and a measure of the quality of higher education. The basic structure of the main outcome variable—quality-adjusted years of higher education (QAYH)—is the following: = × (5) where EYHc represents the expected years of higher education of country c, and QAc represents the average quality of higher education in country c, which has a maximum of 1 and a minimum of m. The minimum is greater than 0 on the assumption even very low-quality higher education has some intrinsic value, even if minimal. QAYH is expressed in years of higher education of maximum quality. 3.1 Expected years of higher education The standard approach for estimating expected years of basic education uses the age-specific enrollment rates over all ages in the 4–18 age range as the main input. The nature of higher education requires a different treatment, for several reasons. First, there is no theoretical age at which higher education is expected to happen. Second, higher education is not always carried out full time; many students combine their studies with part-time employment. Third, the number of years required to obtain a higher education degree varies across disciplines and across countries (the norm in EU countries, after implementation of the Bologna Process, is for initial degrees to take three years; in the Russian Federation, a bachelor’s degrees take four years). The approach adopted in this paper uses the percentage of individuals with a higher education degree at age 30–34 as the measure of educational attainment. To express it in years of education, we assume that a university degree is equivalent to 3.5 years of higher education, to account for differences across disciplines and educational systems. The calculation of expected years of higher education (EYH) is straightforward: 30−34 = × 3.5 (6) where Tertiary attainment corresponds to the share of individuals 30–34 in country c who hold a tertiary degree. 5 3.2 Quality adjustment of higher education attainment Quality adjustment of higher education should be done primarily by measuring the quality of outputs, such as the skill proficiency of university graduates (just as harmonized test score results are used to measure the quality of learning among primary and high school students). However, measures of adult skill proficiency (from the Programme for the International Assessment of Adult Competencies [PIAAC] or Skills Towards Employability and Productivity [STEP] surveys, for example) are available only for a limited set of countries. 1 The ECA-HCI therefore measures the quality of inputs—such as the quality of universities—which are more widely available. However, measures of the quality of universities and adult skill proficiency correlate very well for countries for which both measures are available (see appendix A for a comparison of the input-based quality adjustment presented here and an alternative skill-based quality adjustment). The quality of higher education is calculated under the assumption that a high-quality degree is a degree that makes its holders more productive in the labor market—the working assumption of the broad literature on the effects of college quality on earnings in the United States. Standard ordinary least squares (OLS) estimates of the impact of college quality (usually measured by the average SAT score of admitted students) on earnings show that there is a positive and significant association between them. Given the existence of a selection process into college—high school students decide which colleges to apply to—these estimates may suffer from a substantial selection bias. To address this issue, the literature has followed two approaches. The first is a “selection-on-observables” approach, in which the decision to apply to a given type of college is modeled based on observable variables such as net college costs or high school grade point average (Brewer, Eide, and Ehrenberg 1999; Andrews, Li, and Lovenheim 2016). This approach has confirmed the existence of a positive and significant return of the quality of college education on earnings. The second is a “selection-on-unobservables” approach, in which, rather than modeling college choice, the researcher compares the outcomes of students who were admitted to the same set of colleges but chose to go to different ones (Dale and Krueger 2002, 2014). This approach is a “self-revelation” method, because it assumes that the set of students admitted to a given college share the same “unobservable” characteristics. This method shows that, for the average student, there is no significant effect of college quality on earnings. The effect is significant for minority students and those from poor backgrounds, however. 1 For a comparison of output quality in tertiary education, see Loyalka and others (2019), who compare the computer science skills of computer science undergraduates in their last year in China, India, the Russian Federation, and the United States. 6 The quality-adjustment factor in our study is calculated in the following way: = × × (7) where m corresponds to the productivity of a tertiary degree coming from a “zero-quality” institution; Q corresponds to the average quality score of universities in country c, ranging from 0 to 100; β is a productivity-adjustment factor that transforms the quality score into productivity units; and m is scaled in a way that quality adjustment (QAc) equals 1 if Qc equals 100. The measure of quality corresponds to the information collected by global university rankings. These rankings, published by private, for-profit companies, have grown in number over the years. They are usually based on an underlying score that is usually a weighted average of scores on different aspects of higher education (the volume and quality of research, research influence, the quality of teaching, international outlook, links to industry). These rankings do not include all higher education institutions (universities need to send their information, usually at a cost, to the publishers), and they use different methodologies. Our analysis relies on a combination of several of these ranking, including the scores from the Times Higher Education (THE) ranking; the Quacquarelly Symonds (QS) ranking; the Academic Ranking of World Universities (ARWU, also known as the “Shanghai” ranking); the Center for World University Rankings (CWUR); the U.S. News Global Universities Ranking; and the U-Multirank ranking (a nonnumeric, user- defined ranking). These rankings contain information on 400–1,000 universities in 43 countries in Europe and Central Asia. We generate a country-level average by averaging the scores for all the universities in a given country included in each ranking, yielding six values for each country (one for each ranking source). As detailed later, we normalize each of them, and then take the average of them as the aggregate quality score. University rankings Table 1 describes the six university rankings used in this analysis. The CWUR includes the largest number of universities (2,000); the ARWU/Shanghai includes the smallest number (1,000). The rankings include 385–1,040 higher education institutions in Europe and Central Asia. The total number of countries covered ranges from 63 to 98; the number of countries in Europe and Central Asia ranges from 32 to 43. Five of the six rankings (THE, QS, ARWU, CWUR, and U.S. News rankings) have scores that (theoretically) range from 0 to 100, although no institution included in any of the rankings has a score of 0. The U-Multirank is a nonnumeric, multidimensional, user-defined ranking. To use it, we imputed numeric values (ranging from 0 to 100) to the letter-based scores assigned. The CWUR has the highest minimum score (65.8) and the lowest dispersion (5.07). The ARWU/Shanghai overall score is reported only for the world’s top 100 universities. 7 Given that the six rankings include subcomponents on the quality of research, faculty performance, and reputation, an alternative score can be estimated as the simple average of the scores of those subcomponents—the research, teaching, and citations (RTC) quality score. This score captures the quality of the subcomponents that are common to all the rankings. This calculation is not possible for the CWUR and U.S. News rankings, which do not publish the scores on the subcomponents. Table 1 Descriptions of six systems of university ranking Academic Center for Times Ranking of World U.S. News Higher Quacquarelly World University Global Education Symmonds Universities Rankings Universitie U-Multirank Item (THE) (QS) (ARWU)a (CWUR)b s Ranking (UMR)c Number of 1,397 1,021 1,000 2,000 1,500 1,666 universities included Of which in ECA 540 418 385 708 556 1,041 Number of countries 91 85 63 98 81 92 Of which in ECA 37 35 32 36 36 43 Ranking components covered Research/       innovation on outputs Faculty performance       Internationalization     Reputation      STEM focus  Overall score Global mean 34.57 29.90 37.00 71.64 42.45 59.27 Dispersion 17.07 19.75 12.71 5.07 16.28 14.41 Range 16.4–95.4 10.7–100 26–100 65.8–100 15.5–100 16.7–100 Research, Teaching, and Citations scored Global mean 33.43 30.83 20.96 n.a. n.a. 63.56 Dispersion 17.45 20.00 9.82 n.a. n.a. 16.54 Range 9.3–96.4 10.7–99.9 8.2–92.7 n.a. n.a. 20–100 Note: ECA = Europe and Central Asia; STEM = science, technology, engineering, and mathematics. a. The overall score for the ARWU ranking is published only for the top 100 universities. For the remaining institutions, only the individual subcomponents are published. b. The CWUR publishes only the overall score, not the subcomponent scores. c. The UMR provides a letter-based, not a numeric, score. To estimate a numeric equivalent, the following scale was used: A = 100; B = 75; C = 50, D =2 5, E = 0. The overall score represents the average of the numeric score of all the UMR categories (teaching and learning, research, knowledge transfer, international orientation, and regional engagement). d. The Research, Teaching, and Citations score is composed of the simple average of the components of research, faculty performance, and reputation. The correlation between these rankings is very high. Partial correlations across the rankings for a subset of 98 U.S. universities included in the six rankings range from 0.64 to 0.97 (Table 2). Partial correlations 8 across the country averages for the 54 countries that have at least one university present in all six rankings are also high, ranging from 0.61 to 0.91 (Table 3). Table 2 – Partial correlation across US universities (n=98) Ov. Ov. RTC Ov. Ov. Ov. RTC RTC RTC Ov. Ov. RTC THE QS ARW CWUR US UMR THE QS ARW CWU US UMR U News U R News Overall 1 RTC 1 THE THE Overall 0.9728 1 RTC 0.9544 1 QS QS RTC 0.8762 0.8895 1 RTC 0.8771 0.8735 1 ARWU ARWU Overall 0.9375 0.9492 0.9396 1 Overall 0.9384 0.9437 0.9396 1 CWUR CWUR Overall 0.9395 0.9350 0.9381 0.9620 1 Overall 0.9386 0.9246 0.9381 0.9620 1 US US News News Overall 0.6886 0.7374 0.6486 0.7187 0.7274 1 RTC 0.7230 0.7666 0.6412 0.7246 0.7476 1 UMR UMR Table 3 – Partial correlation across country averages (n=54) Ov. Ov. RTC Ov. Ov. Ov. RTC RTC RTC Ov. Ov. RTC THE QS ARW CWUR US UMR THE QS ARW CWU US UMR U News U R News Overall 1 RTC 1 THE THE Overall 0.9044 1 RTC 0.8833 1 QS QS RTC 0.8514 0.8587 1 RTC 0.8436 0.8533 1 ARWU ARWU Overall 0.8741 0.8443 0.8461 1 Overall 0.8819 0.8427 0.8461 1 CWUR CWUR Overall 0.9272 0.8220 0.7975 0.8961 1 Overall 0.9144 0.7962 0.7975 0.8961 1 US US News News Overall 0.7453 0.7711 0.6977 0.7122 0.6387 1 RTC 0.7354 0.7231 0.7210 0.7125 0.6138 1 UMR UMR A positive correlation also exists between the quality scores and the income level of countries (see Figure 1). Singapore is ranked as the country with the highest quality score in the THE, QS and ARWU rankings, while for the CWUR ranking the highest ranked country is the Netherlands. 9 Figure 1 – University rankings (quality score) and income level Note: in red are marked the countries of Europe and Central Asia To create an aggregate quality score that combines the information from the six rankings, we first code as 0 the score for a country that is not present in the ranking (except for the CWUR ranking, for which we use a value of 60, given that the minimum score recorded in that ranking is 66.5). The scores for each ranking are then normalized to have a mean of 0 and a standard deviation of 1. The overall score is used for the THE, QS, CWUR, U.S. News, and U-Multirank rankings; the RTC score is used for the ARWU. The simple average of the six standardized scores is then rescaled to a 0–100 range for presentational purposes. This procedure ranks countries in terms of the average quality of its universities, ignoring the distribution of students across universities. Given that this information is not available at a global scale, the simple average is used. 10 Figure 2 plots the values of the aggregate quality score by country and income level. Only countries that are present in at least one of the six rankings are included. 2 The correlation between income level and the aggregate quality score is particularly steep for Europe and Central Asia. Figure 2. Correlation between aggregate quality score of universities and country income level Note: Only countries present in at least one of the six rankings are included. Red points indicate countries in Europe and Central Asia. Estimation of the quality-adjustment factor To estimate the productivity effect of university quality (parameters β and m in equation 7), we rely on a cohort-college-level data set for 294 U.S. colleges. Focusing on the U.S. data allows us to control for parental income, one of the key drivers of individual income. The data set comes from the Mobility Report Cards constructed by Chetty and others (2017), which combines college and administrative data that link the parental and post-college earnings of about 28.1 million students born between 1980 and 1991 for 2,463 colleges. The data set consists of cohort-college observations—that is, observations of the average characteristics of students born in a given year who studied at a given college. For each observation, the data set includes the students’ average annual earnings in 2014 and the average parental earnings when the cohort was age 15–19. The data set also includes a series of college-level variables, such as the average 2 In that countries have a value of zero in the quality score Qc, the quality adjustment factor QAc has a value of m: having a tertiary degree from these countries has an intrinsic value of m but no additional quality premium. 11 attendance costs, instructional expenditure, and percentage of students in each type of major. We match this data set with the six university rankings. Among U.S. higher education institutions, 294 are present in at least one of the rankings, and 108 are present in all four. The simple OLS regression estimated is the following: 2014 log () ,, = + + 1, log (),, + 2, + 3, _ + ,, (8) where the dependent variable is the annual average log earnings in 2014 of the cohort born in year b of gender g that went to college c. The main regressor of interest is Q, the quality measure based on the six rankings for college c. Coefficient β is the productivity effect of quality; it is used as the quality-adjustment factor in equation (7) which feeds into the ECA-HCI. Other regressors are the log parental earnings of the cohort born in year b of gender g that went to college c when the individuals were 15–19; the age of cohort b in 2014; and percentage of STEM majors in college c in year 2000 (included to control for the STEM wage premium). Standard errors are clustered at the college level. Table 4 provides the results for the aggregate quality score derived from the combination of the six rankings, shown for the sample of universities that are present in at least one of the rankings (323 universities in total) and for the common sample of 98 universities that are present in all the rankings. Table 5 summarizes the values of β and m (the implied productivity of a “zero-quality” institution) that arise from the results of the OLS estimations of equation (8), focusing only on values that refer to both genders. Full results are available in Table A.1 in the appendix. Table 4. Ordinary least squares estimates of aggregate quality scores of universities Log annual earnings in 2014 Full sample Common sample Both genders Men Women Both genders Men Women Aggregate quality score 0.0024*** 0.0031*** 0.0016*** 0.0044*** 0.0052*** 0.0036*** (0.004) (0.0004) (0.0004) (0.0009) (0.0010) (0.0008) Log parental earnings 0.2986*** 0.3142*** 0.2646*** 0.3202*** 0.3597*** 0.2543*** (0.0136) (0.0150) (0.0134) (0.0248) (0.0283) (0.0225) Age 0.1074*** 0.1237*** 0.0894*** 0.1157*** 0.1295*** 0.0979*** (0.0013) (0.0014) (0.0013) (0.0023) (0.0024) (0.0022) STEM majors in 0.0058*** 0.0053*** 0.0049*** 0.0046*** 0.0043*** 0.0032*** college (0–100) (0.0005) (0.0005) (0.0006) (0.0008) (0.0008) (0.0010) (percent) Constant 3.8250*** 3.2798*** 4.6636*** 3.2758*** 2.4979*** 4.5139*** (0.1679) (0.1881) (0.1606) (0.3061) (0.3469) (0.2707) Observations 3,784 3,689 3,738 1,159 1,159 1,156 Number of colleges 323 315 321 98 98 98 Note: The common sample is composed of universities that are present in all six rankings. Clustered standard errors at the college level are in parentheses. STEM = science, technology, engineering, and mathematics. * p < 0.10, ** p < 0.05, *** p < 0.01. 12 Table 5. Parameters of the quality-adjustment factor used to assess universities U.S. U-Multirank Aggregate quality THE QS ARWU CWUR News score (overall) Overall RTC Overall RTC RTC Overall Overall Overall RTC All Common sample (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) β 0.0032 0.0031 0.0027 0.0024 0.0045 0.0073 0.0019 0.0040 0.0032 0.0024 0.0044 m 0.726 0.733 0.763 0.787 0.638 0.747 0.826 0.668 0.728 0.787 0.647 Note: ARWU = Academic Ranking of World Universities; CWUR = Center for World University Rankings; QS = Quacquarelly Symonds; RTC = research, teaching, and citations; THE = Times Higher Education. To estimate the ECA-HCI, we use the values estimated from the use of the aggregate quality score in the extended sample (Table 5, column 10). These values can be understood as a conservative estimate of the productivity effects of quality, as the estimates from the sample of universities present in the six rankings (Table 5, column 11) imply a larger effect. The parameters are derived from the implied differences in the wages of graduates of a low-quality university compared with those of a high-quality university in the United States. This implied wage differential may be even higher when comparing a low-quality university in a given country with a high-quality university in another country. Interpretation of the results emerging from the use of this quality-adjustment factor needs to take these limitations into account. 3.3 Quality-adjusted years of higher education Based on the estimates of the previous paragraphs, the detailed calculation formula for the quality-adjusted years of higher education (QAYH) is as follows: 30−34 = × 3.5 × 0.787 × 0.0024× (8) where Q is the aggregate quality score for higher education for country c. There is also positive association between quality-adjusted years of higher education and income level (see Figure 3). 13 Figure 3 – Quality-adjusted years of higher education and income level Note: Countries in blue are in Europe and Central Asia. Countries in red are high income countries from other regions. 3.4 Contribution of education to relative productivity in the ECA-HCI To calculate the education component of the ECA-HCI, we need to establish the returns to one additional year of tertiary education. We rely on the evidence presented by Montenegro and Patrinos (2014), who suggest that an average return of an additional year of tertiary education is 0.152. For basic education, as stated before, we will rely on the same productivity parameter as the original HCI – a value of 0.08 for any additional year of basic education. The benchmarks for full education are set at 14 years of basic education and 3.5 years of higher education. Country and subregion estimates are presented in table 6. Figure 4 plots the education component of the ECA-HCI. The distance between a given value and 1 indicates the productivity lost as a result of the average level of education falling short of the benchmark There is a positive association between a country’s income level and the contribution of education to relative productivity, although the association is looser at lower income levels. 14 Figure 4. Contribution of education to relative productivity (ECA-HCI) Table 6 Education component of the Europe and Central Asia extension of the Human Capital Index (ECA-HCI) Learning- Share of Aggregate Quality- adjusted population higher adjusted years of 30–34 with education years of Education Education basic tertiary quality higher component, component, Subregion/country education degree score education ECA-HCI HCI Central Asia 8.8 0.213 2.5 0.59 0.424 0.659 Kazakhstan 9.1 0.344 9.3 0.97 0.461 0.677 Kyrgyz Republic 8.7 0.295 0.81 0.433 0.652 Tajikistan 6.8 0.224 0.62 0.362 0.561 Uzbekistan 9.1 0.121 0.34 0.419 0.678 Central Europe and Baltic 10.4 0.392 22.3 1.14 0.526 0.750 countries Bulgaria 8.7 0.324 21.6 0.94 0.443 0.654 Croatia 10.4 0.281 23.8 0.82 0.501 0.753 Czech Republic 11.1 0.357 25.1 1.04 0.547 0.794 Estonia 11.7 0.480 29.7 1.42 0.607 0.833 Hungary 10.3 0.296 23.5 0.86 0.497 0.742 Latvia 11.0 0.440 17.3 1.26 0.559 0.785 Lithuania 11.0 0.567 21.8 1.65 0.592 0.785 Poland 11.4 0.485 22.1 1.41 0.590 0.811 Romania 8.4 0.298 20.6 0.86 0.427 0.637 Slovak Republic 9.8 0.366 20.6 1.06 0.493 0.715 Slovenia 11.4 0.411 25.9 1.20 0.572 0.810 15 Eastern Europe 9.9 0.525 16.3 1.50 0.534 0.723 Belarus 10.8 0.421 22.1 1.22 0.547 0.773 Moldova 8.3 0.351 5.8 0.98 0.432 0.633 Ukraine 9.9 0.560 15.9 1.60 0.539 0.719 Northern Europe 11.4 0.512 37.3 1.54 0.605 0.814 Denmark 11.1 0.579 42.8 1.77 0.609 0.793 Finland 11.7 0.428 35.1 1.28 0.596 0.835 Iceland 10.7 0.537 30.4 1.59 0.575 0.769 Norway 11.2 0.502 32.8 1.50 0.591 0.801 Sweden 11.6 0.524 38.1 1.58 0.616 0.824 Russian Federation 10.9 0.610 25.9 1.79 0.601 0.780 South Caucasus 8.2 0.299 8.5 0.84 0.421 0.630 Armenia 8.0 0.303 9.4 0.85 0.414 0.619 Azerbaijan 8.3 0.254a 6.4 0.71 0.414 0.633 Georgia 8.3 0.417 13.1 1.19 0.445 0.632 Southern Europe 10.5 0.341 29.8 1.01 0.518 0.756 Cyprus 10.9 0.558 27.3 1.64 0.589 0.781 Greece 10.0 0.446 25.5 1.30 0.519 0.724 Italy 10.5 0.271 31.6 0.80 0.500 0.753 Malta 10.2 0.341 16.0 0.98 0.502 0.737 Portugal 11.3 0.327 28.5 0.96 0.548 0.806 Spain 10.5 0.406 28.7 1.20 0.533 0.757 Turkey 9.2 0.275 22.2 0.80 0.453 0.683 Western Balkans 8.8 0.285 13.3 0.81 0.442 0.664 Albania 9.0 0.235 6.9 0.66 0.434 0.668 Bosnia and Herzegovina 7.8 0.210 5.5 0.59 0.391 0.609 Kosovo 7.9 0.612 Montenegro 8.9 0.340 1.5 0.94 0.451 0.665 North Macedonia 7.3 0.299 7.3 0.84 0.390 0.585 Serbia 9.8 0.333 22.6 0.97 0.485 0.712 Western Europe 11.3 0.459 36.2 1.38 0.583 0.803 Austria 10.9 0.472 32.1 1.40 0.568 0.781 Belgium 11.2 0.489 42.3 1.49 0.588 0.798 France 11.3 0.470 33.1 1.40 0.584 0.804 Germany 11.0 0.340 35.5 1.02 0.541 0.789 Ireland 11.6 0.597 33.2 1.78 0.635 0.825 Luxembourg 9.8 0.497 28.1 1.46 0.524 0.714 Netherlands 11.5 0.550 46.8 1.70 0.624 0.821 Switzerland 10.9 0.512b 45.9 1.58 0.584 0.782 United Kingdom 11.5 0.550 35.9 1.65 0.620 0.821 ECA (country average) 10.1 0.403 23.0 1.18 0.520 0.736 ECA (population-weighted 10.4 0.424 26.2 1.25 0.539 0.755 average) 16 Sources: Attainment data were calculated from the European Union Statistics on Income and Living Conditions and household surveys. Learning-adjusted years of basic education (LAYS) were obtained from the HCI database. Note: For the average standardized quality score for higher education, the quality scores from each of the six university rankings (the Times Higher Education, the Quacquarelly Symonds, Academic Ranking of World Universities, the Center for World University Rankings, the U.S. News Global Universities Ranking, and U-Multirank) are first standardized to a global mean of 0 and a standard deviation of 1 and then averaged for every country. For presentational purposes, this value is then rescaled to range from 0 to 100. A value of 0 for the quality measure implies that no university in that country appears in any of the six university rankings. The education component of the original HCI was updated with PISA 2018 results or the latest available data. HCI = Human Capital Index; ECA-HCI = Europe and Central Asia extension of the HCI. — Not available. a. Based on population age 25 and older. b. Based on population 25–34. 4. Health Component The health component of the HCI seeks to measure the productivity losses associated with poor health that a child born today will face later in life as an adult. The original HCI calculates this component based on two variables: the child stunting rate and the adult survival rate (the chance that a 15-year-old survives to age 60). These variables are understood to be good proxies for unobserved latent health status in a global context. Their effects on productivity are measured by the returns to adult height. The ECA-HCI takes a different approach. It starts by assuming that good health means the absence of disease and bad health means the presence of disease. To measure latent health status, the ECA-HCI focuses on the factors that may cause disease. A low prevalence of these risk factors implies a lower disease burden; a high prevalence could imply a higher disease burden. The risk factors that are relevant as indirect measures of latent health status depend on the types of disease prevalent in each context. Smith and Nguyen (2013) show that in Europe and Central Asia, cardiovascular disease, followed by external causes (mainly alcohol- related road traffic injuries), explains most of the differences in adult life expectancy. Data from the COVID-19 pandemic also show that people with underlying cardiovascular conditions have a higher mortality rate than people without them (Wu and McGoogan 2020; Zhou and others 2020). In view of these findings, the ECA-HCI uses the prevalence of three health risk factors associated with cardiovascular disease: obesity, tobacco smoking, and heavy alcohol consumption. The higher the prevalence of these risk factors, the higher the probability of disease and the worse the health status. The prevalence of these risk factors increases the probability of suffering from noncommunicable diseases and increases the mortality and morbidity consequences of some infectious diseases like COVID-19. The health benchmark in the ECA-HCI with which countries are compared is zero prevalence of obesity, smoking, and heavy drinking. The impact on productivity of specific health conditions is difficult to estimate. There is more evidence on the productivity effects associated with the risk factors behind such health conditions. The literature has quantified the effects on productivity of obesity, tobacco smoking, and heavy drinking, making it possible 17 to incorporate their prevalence directly into the ECA-HCI without the intermediating factor of adult height, as in the original version of the index. Focusing only on risk factors has its limitations, however. Between risk factors and morbidity lies a mediating institutional factor: health care systems. The capacity of health care systems to manage the consequences of increased risk factors—and the diseases associated with them—ultimately determines whether that increased risk ends in increased morbidity and, eventually, mortality. Good health care systems strongly alleviate the morbidity and mortality consequences of the increased prevalence of risk factors. To account for the effects of health care systems, the model uses a health outcome measure as a proxy for latent health status—the child stunting and adult survival rates used in the original HCI. The health component of the ECA-HCI uses the average of a risk factor–based proxy of health status and an outcome- based proxy. The productivity effects of child stunting and adult survival rates are retained, as in the original HCI. The health component of the ECA-HCI has the following basic formulation: (−∗ )+ (−∗ ) − ℎ = 2 (9) where γRF is the productivity effect associated to the prevalence of risk factors RF; RF* is the benchmark rate of zero prevalence of risk factors; and γO is the productivity effect of health outcomes O, with the benchmark of “full” health outcomes being O*. For risk factors, the ECA-HCI uses the share of non-obese adults (NOB), the share of nonsmokers among adults (NSM), and the share of adults not reporting heavy drinking (NAL). The productivity effects of these risk factors (γOB, γSM, γAL) are assumed to be additive. 3 For health outcomes, the ECA-HCI uses the adult survival rate (ASR) and the share of children not stunted (NSTNT). As in the original HCI, these rates are intended to proxy the same variable: latent health status. Their productivity effects (γASR, γSTNT) are therefore averaged. The equation for the health component is the following: [ (−1)+ (−1)+ (−1)]+[ (−1)+ (−1)]/2 − ℎ = 2 (10) The values of γSTNT and γASR, the productivity effects associated with child stunting the adult survival rate, are kept as in the original HCI. They are derived from the correlation of these rates with adult height, for which the literature provides reliable microeconometric estimations of productivity. These values are assumed to be 0.35 for γSTNT and 0.65 for γASR. Adult survival rates are widely available; child stunting rates 3 Perfectly additive productivity effects imply that the productivity effect of smoking and obesity (combined) is simply the summation of the productivity effect of smoking and the productivity effect of obesity. This figure can be understood as an upper-bound estimation of the combined productivity effects of risk factors. 18 are available only for a few countries in the region. For countries for which estimates of child stunting are not available, only the adult survival rate is used to estimate the outcome-based productivity proxy. A literature review was carried out to obtain estimates of the productivity effects of the prevalence of the risk factors (see appendix B). The median values for all the average effects found was chosen as the parameter for use in the ECA-HCI. These values are 0.0993 for obesity (γOB), 0.096 for smoking (γSM), and 0.1995 for heavy drinking (γAL). These values represent the negative productivity effects associated with each risk factor. The prevalence of the three health risk factors among the adult population across Europe and Central Asia is plotted in figure 5 in comparison with country income levels. Figure 5 – Prevalence of health risk factors among adult population Sources: European Health Interview Survey 2014 and World Health Organization. Country and subregional estimates of the health component are presented in table 7. Figure 6 plots the values of the health component with respect to countries’ income level. In contrast to the education 19 component, there is no clear correlation between income and the contribution of health status to relative productivity. Figure 6. Contribution of health to relative productivity in Europe and Central Asia Note: Productivity figures are from the Europe and Central Asia extension of the Human Capital Index (ECA-HCI). Table 7 Health component of the Europe and Central Asia extension of the Human Capital Index (ECA-HCI) Obese adult Heavy Current Adult Children under Health Health population episodic smokers survival 5 not stunted component, component, Subregion/country (%) drinkers (%) (%) rate (%) ECA-HCI HCI Central Asia 16.6 11.5 17.8 0.859 88.9 0.941 0.937 Kazakhstan 21.3 19.9 24.3 0.845 92.0 0.928 0.937 Kyrgyz Republic 15.4 11.1 26.4 0.849 88.2 0.936 0.933 Tajikistan 12.6 7.9 18.8 0.871 82.5 0.942 0.930 Uzbekistan 15.3 7.9 12.3 0.866 89.2 0.949 0.939 Central Europe and Baltic 15.9 19.3 27.3 0.890 – 0.928 0.933 Countries Bulgaria 14.4 17.1 34.8 0.866 93.0 0.934 0.946 Croatia 19.0 10.9 28.7 0.917 – 0.941 0.948 Czech Republic 18.8 14.9 28.7 0.922 – 0.939 0.951 Estonia 19.6 23.3 27.6 0.897 – 0.924 0.936 Hungary 20.6 8.3 27.5 0.880 – 0.932 0.925 Latvia 21.3 19.2 29.5 0.844 – 0.910 0.904 Lithuania 16.6 20.1 25.0 0.844 – 0.913 0.903 20 Poland 16.7 17.4 26.1 0.894 – 0.930 0.934 Romania 9.1 34.9 25.7 0.878 – 0.913 0.924 Slovak Republic 15.9 12.8 29.5 0.898 – 0.934 0.936 Slovenia 18.6 19.0 24.2 0.935 – 0.941 0.959 Eastern Europe 25.8 22.0 26.9 0.822 – 0.901 0.893 Belarus 26.6 28.2 26.2 0.853 93.6 0.903 0.909 Moldova 20.1 28.6 24.2 0.836 – 0.921 0.937 Ukraine 26.1 20.2 27.3 0.815 – 0.899 0.886 Northern Europe 14.4 31.5 18.8 0.941 – 0.936 0.962 Denmark 14.4 37.4 20.9 0.932 – 0.926 0.957 Finland 17.8 33.9 19.2 0.930 – 0.928 0.956 Iceland 19.0 25.7 18.8 0.955 – 0.943 0.971 Norway 12.6 44.0 20.1 0.945 – 0.925 0.965 Sweden 13.4 20.4 16.7 0.950 – 0.950 0.968 Russian Federation 25.0 38.8 30.3 0.804 – 0.879 0.880 South Caucasus 20.8 11.1 23.1 0.876 – 0.934 0.930 Armenia 20.9 11.5 24.5 0.886 90.6 0.941 0.948 Azerbaijan 19.9 8.2 20.8 0.882 82.2 0.939 0.933 Georgia 23.3 18.5 28.0 0.853 – 0.913 0.909 Southern Europe 13.6 8.2 24.3 0.947 – 0.957 0.966 Cyprus 13.1 5.2 29.1 0.952 – 0.960 0.969 Greece 16.9 10.3 32.6 0.933 – 0.945 0.957 Italy 10.5 6.6 22.7 0.953 – 0.963 0.970 Malta 25.2 19.2 24.1 0.951 – 0.943 0.969 Portugal 16.1 10.2 20.0 0.933 – 0.952 0.957 Spain 16.2 9.3 25.3 0.946 – 0.954 0.966 Turkey 19.8 4.3 32.5 0.911 94.0 0.952 0.961 Western Balkans 22.5 27.9 35.0 0.906 92.4 0.925 0.957 Albania 22.3 22.9 28.9 0.929 88.7 0.933 0.958 Bosnia and Herzegovina 19.4 22.7 38.1 0.914 91.1 0.930 0.957 Kosovo – – – 0.906 – – 0.941 Montenegro 24.9 26.9 35.4 0.906 90.6 0.923 0.954 North Macedonia 23.9 26.5 35.0 0.909 95.1 0.928 0.962 Serbia 23.5 32.9 36.0 0.893 94.0 0.919 0.956 Western Europe 16.5 29.7 23.0 0.933 – 0.932 0.957 Austria 14.3 18.7 30.0 0.937 – 0.941 0.960 Belgium 13.7 27.5 23.0 0.931 – 0.935 0.956 Germany 14.7 36.0 28.3 0.926 – 0.922 0.953 France 16.4 33.0 21.7 0.931 – 0.929 0.956 Ireland 28.1 32.3 22.0 0.944 – 0.928 0.964 Luxembourg 15.1 34.5 20.5 0.942 – 0.932 0.963 Netherlands 12.9 31.6 25.2 0.946 – 0.935 0.966 Switzerland 11.3 15.9 27.1 0.954 – 0.952 0.970 21 United Kingdom 20.1 22.1 17.3 0.933 – 0.940 0.958 ECA (country average) 18.0 21.1 25.9 0.904 90.3 0.932 0.945 ECA (population-weighted average) 18.4 22.5 25.6 0.894 91.4 0.927 0.938 Source: Data on obesity, smoking, and alcohol consumption are from the European Health Interview Survey, Health Equity and Financial Protection Indicators, and the World Health Organization. The ECA average for the share of children not stunted is calculated based on countries for which data are available only. Note: HCI = Human Capital Index; ECA-HCI = Europe and Central Asia extension of the HCI. — Not available. a. Includes consumption of smokeless tobacco. 5. Estimation of the ECA-HCI The ECA-HCI is the product of three components: − = × × ℎ The three components are defined as follows: 1 − 5 ≡ 1 ≡ 0.08(−14)+0.152(−3.5) [0.0993(−1)+0.096(−1)+0.1995(−1)]+[0.65(−1)+0.35(−1)]/2 ℎ ≡ 2 . The estimates of the ECA-HCI in table 8 show that countries in the region can achieve large increases in their long-run productivity if they reduce the distance between the expected educational attainment and adult health status of children born today and the benchmarks of complete education and full health. The average country for which the ECA-HCI is calculated has a value of 0.481, meaning that children born today in the average country in the region will be almost half as productive as they would have had they reached the benchmark of complete education and full health (14 years of basic education; 3.5 years of higher education; no obesity, tobacco smoking, or heavy drinking; no statistically significant child stunting; and 100 percent adult survival rate to age 60). The correlation between income levels and the ECA-HCI is positive, as it is for the original HCI (figure 7). Figure 7. Estimates of ECA-HCI and country income levels 22 Note: Figures are based on the Europe and Central Asia extension of the Human Capital Index extension of (ECA- HCI). The value of the ECA-HCI is consistently below that of the original HCI, because the full education benchmark of the ECA-HCI includes higher education. However, there is considerable correlation between the two values (figure 8), although some re-ranking occurs. Like the original HCI, the ECA-HCI is measured with some imprecision, so small differences across countries do not represent meaningful differences in education and health environments. Figure 8. Correlation between the original HCI and ECA-HCI 23 Note: Figures are based on the Europe and Central Asia extension of the Human Capital Index extension of (ECA- HCI) and the HCI 2020 Update. Table 8 Full estimates of the Europe and Central Asia extension of the Human Capital Index (ECA- HCI) Probability of survival Education Health ECA- Subregion/country to age 5 component component HCI HCI Central Asia 0.980 0.424 0.941 0.391 0.606 Kazakhstan 0.990 0.461 0.928 0.424 0.629 Kyrgyz Republic 0.981 0.433 0.936 0.398 0.597 Tajikistan 0.965 0.362 0.942 0.330 0.504 Uzbekistan 0.979 0.419 0.949 0.389 0.623 Central Europe and Baltic countries 0.995 0.526 0.928 0.486 0.697 Bulgaria 0.993 0.443 0.934 0.411 0.614 Croatia 0.995 0.501 0.941 0.469 0.710 Czech Republic 0.997 0.547 0.939 0.511 0.752 Estonia 0.997 0.607 0.924 0.559 0.777 Hungary 0.996 0.497 0.932 0.461 0.683 Latvia 0.996 0.559 0.910 0.506 0.707 Lithuania 0.996 0.592 0.913 0.538 0.706 Poland 0.996 0.590 0.930 0.546 0.753 Romania 0.993 0.427 0.913 0.387 0.584 Slovak Republic 0.994 0.493 0.934 0.458 0.665 24 Slovenia 0.998 0.572 0.941 0.537 0.775 Eastern Europe 0.992 0.534 0.901 0.477 0.640 Belarus 0.997 0.547 0.903 0.492 0.700 Moldova 0.984 0.432 0.921 0.391 0.584 Ukraine 0.991 0.539 0.899 0.480 0.631 Northern Europe 0.997 0.605 0.936 0.564 0.781 Denmark 0.996 0.609 0.926 0.562 0.755 Finland 0.998 0.596 0.928 0.552 0.796 Iceland 0.998 0.575 0.943 0.541 0.745 Norway 0.997 0.591 0.925 0.545 0.771 Sweden 0.997 0.616 0.950 0.583 0.795 Russian Federation 0.993 0.601 0.879 0.525 0.681 South Caucasus 0.983 0.421 0.934 0.386 0.576 Armenia 0.988 0.414 0.941 0.385 0.579 Azerbaijan 0.978 0.414 0.939 0.381 0.578 Georgia 0.990 0.445 0.913 0.402 0.569 Southern Europe 0.997 0.518 0.957 0.494 0.728 Cyprus 0.998 0.589 0.960 0.564 0.756 Greece 0.996 0.519 0.945 0.488 0.690 Italy 0.997 0.500 0.963 0.480 0.728 Malta 0.993 0.502 0.943 0.470 0.709 Portugal 0.996 0.548 0.952 0.520 0.769 Spain 0.997 0.533 0.954 0.507 0.728 Turkey 0.989 0.453 0.952 0.426 0.649 Western Balkans 0.993 0.442 0.925 0.406 0.631 Albania 0.991 0.434 0.933 0.401 0.634 Bosnia and Herzegovina 0.994 0.391 0.930 0.362 0.580 Kosovo 0.985 0.567 Montenegro 0.997 0.451 0.923 0.415 0.633 North Macedonia 0.990 0.390 0.928 0.359 0.557 Serbia 0.994 0.485 0.919 0.443 0.677 Western Europe 0.996 0.583 0.932 0.541 0.765 Austria 0.996 0.568 0.941 0.533 0.747 Belgium 0.996 0.588 0.935 0.548 0.760 France 0.996 0.584 0.922 0.537 0.763 Germany 0.996 0.541 0.929 0.501 0.751 Ireland 0.996 0.635 0.928 0.587 0.793 Luxembourg 0.998 0.524 0.932 0.487 0.686 Netherlands 0.996 0.624 0.935 0.581 0.790 Switzerland 0.996 0.584 0.952 0.553 0.756 United Kingdom 0.996 0.620 0.940 0.580 0.783 Simple average 0.993 0.520 0.932 0.481 0.691 Population-weighted average 0.993 0.539 0.927 0.496 0.704 25 Source: Authors’ calculations. Note: HCI = Human Capital Index; ECA-HCI = Europe and Central Asia extension of the Human Capital Index. Gender disaggregation of ECA-HCI Like the original HCI, the ECA-HCI can be disaggregated by gender. The values of learning-adjusted years of schooling can be disaggregated by gender in terms of quantity (expected years of basic education) and quality (test score performance); the values of QAYH can be disaggregated by gender in quantity (expected years of higher education) but not by quality, as there is no gender variation in the quality measure used for higher education (university rankings). The prevalence of adult risk factors (obesity, smoking, and heavy drinking) is available for men and women for almost all countries in the region. The results can be disaggregated by gender for 38 countries (table 9). For the average country, the value of the ECA-HCI is 0.459 for men and 0.517 for women. In all countries, the value is lower for men than women (figure 9). The gender gap is largest in Finland and Latvia (about 11 percentage points) and smallest in Uzbekistan and Turkey (1 percentage point or below). Figure 9. Gender-disaggregated values of ECA-HCI 26 Note: Figures are based on the Europe and Central Asia extension of the Human Capital Index extension of (ECA- HCI). Table 9 Gender-disaggregated estimates of the Europe and Central Asia extension of the Human Capital Index (ECA-HCI) Probability of Education survival to age 5 component Health component ECA-HCI Subregion/country Men Women Men Women Men Women Men Women Central Asia 0.978 0.983 0.433 0.434 0.920 0.961 0.390 0.411 Kazakhstan 0.989 0.991 0.449 0.474 0.900 0.956 0.399 0.449 Kyrgyz Republic 0.979 0.983 0.424 0.442 0.909 0.962 0.378 0.418 Tajikistan 0.961 0.969 – – – – – – Uzbekistan 0.976 0.982 0.426 0.411 0.933 0.964 0.388 0.389 Central Europe and Baltic 0.995 0.995 0.503 0.549 0.909 0.957 0.456 0.524 countries Bulgaria 0.992 0.994 0.431 0.457 0.914 0.953 0.391 0.432 Croatia 0.995 0.996 0.479 0.525 0.921 0.961 0.439 0.502 Czech Republic 0.996 0.997 0.523 0.570 0.921 0.957 0.480 0.544 Estonia 0.997 0.998 0.579 0.641 0.894 0.954 0.516 0.610 Hungary 0.995 0.996 0.487 0.509 0.910 0.953 0.441 0.483 Latvia 0.996 0.996 0.521 0.600 0.873 0.945 0.453 0.565 Lithuania 0.996 0.996 0.562 0.625 0.873 0.952 0.488 0.592 Poland 0.995 0.996 0.557 0.622 0.923 0.964 0.512 0.597 Romania 0.992 0.993 0.418 0.436 0.877 0.949 0.364 0.411 Slovak Republic 0.994 0.995 0.470 0.518 0.911 0.958 0.426 0.494 Slovenia 0.998 0.998 0.533 0.614 0.923 0.959 0.491 0.588 Eastern Europe 0.991 0.993 0.517 0.551 0.860 0.942 0.441 0.515 Belarus 0.996 0.997 0.534 0.560 0.863 0.943 0.459 0.526 Moldova 0.982 0.986 0.419 0.443 0.888 0.953 0.366 0.416 Ukraine 0.990 0.992 0.521 0.558 0.857 0.940 0.443 0.520 Northern Europe 0.997 0.997 0.570 0.643 0.921 0.950 0.523 0.610 Denmark 0.995 0.996 0.575 0.647 0.912 0.940 0.522 0.606 Finland 0.998 0.998 0.553 0.644 0.907 0.949 0.500 0.611 Iceland 0.998 0.998 0.536 0.621 0.934 0.953 0.500 0.591 Norway 0.997 0.998 0.557 0.630 0.911 0.939 0.506 0.590 Sweden 0.997 0.998 0.584 0.648 0.939 0.962 0.546 0.622 Russian Federation 0.992 0.994 0.582 0.623 0.840 0.917 0.485 0.568 South Caucasus 0.981 0.985 0.417 0.426 0.908 0.958 0.371 0.402 Armenia 0.986 0.989 0.401 0.427 0.912 0.967 0.361 0.409 Azerbaijan 0.976 0.981 0.416 0.412 0.921 0.957 0.374 0.387 Georgia 0.989 0.991 0.431 0.461 0.871 0.954 0.371 0.436 Southern Europe 0.997 0.997 0.503 0.534 0.945 0.968 0.473 0.515 Cyprus 0.997 0.998 0.574 0.604 0.944 0.974 0.541 0.586 Greece 0.995 0.996 0.499 0.540 0.929 0.961 0.462 0.517 27 Italy 0.997 0.997 0.481 0.519 0.953 0.969 0.457 0.502 Malta 0.992 0.994 0.478 0.530 0.930 0.956 0.441 0.504 Portugal 0.996 0.997 0.529 0.568 0.933 0.969 0.492 0.549 Spain 0.997 0.997 0.524 0.542 0.942 0.967 0.492 0.523 Turkey 0.989 0.990 0.453 0.452 0.939 0.965 0.421 0.432 Western Balkans 0.993 0.994 0.429 0.460 0.904 0.947 0.385 0.432 Albania 0.991 0.992 – – 0.910 0.957 – – Bosnia and Herzegovina 0.994 0.995 0.378 0.405 0.910 0.951 0.341 0.384 Kosovo 0.983 0.988 – – – – – – Montenegro 0.997 0.998 0.443 0.458 0.904 0.943 0.400 0.431 North Macedonia 0.989 0.991 0.375 0.407 0.909 0.949 0.337 0.383 Serbia 0.994 0.995 0.469 0.502 0.897 0.941 0.418 0.470 Western Europe 0.996 0.996 0.571 0.595 0.914 0.949 0.520 0.563 Austria 0.996 0.997 0.564 0.572 0.927 0.956 0.520 0.545 Belgium 0.996 0.997 0.567 0.611 0.919 0.950 0.519 0.579 France 0.996 0.996 0.558 0.610 0.897 0.948 0.499 0.576 Germany 0.996 0.997 0.537 0.545 0.913 0.945 0.488 0.514 Ireland 0.996 0.997 0.613 0.655 0.917 0.951 0.560 0.621 Luxembourg 0.997 0.998 0.501 0.547 0.916 0.948 0.458 0.517 Netherlands 0.996 0.997 0.601 0.649 0.916 0.955 0.548 0.618 Switzerland 0.996 0.996 0.575 0.593 0.942 0.962 0.539 0.568 United Kingdom 0.995 0.996 0.615 0.625 0.927 0.952 0.568 0.592 Simple average 0.993 0.994 0.507 0.545 0.910 0.954 0.459 0.517 Population-weighted average 0.993 0.994 0.527 0.556 0.905 0.950 0.473 0.524 Source: Authors’ calculations. Note: – Not available. Uncertainty intervals of ECA-HCI The components of the ECA-HCI are measured with some error; just as in the original HCI, an uncertainty interval can be calculated to provide a measure of the precision of the estimates. This uncertainty interval is not a statistical estimation but rather a calculation of the ECA-HCI under worst- or best-case scenarios. The worst-case scenario indicates that all the components take the lower-bound values; the best-case scenario indicates that all the components take the upper-bound values. As Kraay (2019) points out, this approach is conservative, equivalent to assuming that the measurement error is highly correlated across components. The variables for which lower- and upper-bound values are available are the probability of survival to age five; quality-adjustment factors for basic education (harmonized learning outcomes) and higher education (aggregate quality score); the prevalence of adult health risk factors (obesity, smoking, and heavy drinking); the adult survival rate; and the share of stunted children. 28 For the probability of survival to age five, harmonized learning outcomes, the adult survival rate, and the share of stunted children, we use the same bounds as in the original HCI (for details, see Kraay 2019). For the aggregate quality score for higher education, we use as bounds the maximum and minimum values for each country across the six university rankings (after rescaling the CWUR ranking to 0–100). For the adult health risk factors, the determination of the bounds depends on the data source. For countries whose values are sourced from the European Health Interview Survey, the bounds represent the limits of the 95 percent confidence interval, as detailed in the European Health Interview Survey round 2 quality report (Eurostat, 2018). For countries whose values are sourced from the World Health Organization, the bounds are that institution’s low and high estimates. The ECA-HCI values range from 0.31 to 0.60 (see table 10). The median size of the uncertainty intervals is about 0.025—very similar to that of the original HCI (0.030). For some countries with less precise component data, the interval can range up to 0.076. Figure 10 plots the uncertainty intervals of the ECA- HCI. Figure 10. Uncertainty intervals for ECA-HCI Note: ECA-HCI estimate in blue. Grey lines indicate the upper and lower bounds estimates. 29 Table 10 Uncertainty intervals for the Europe and Central Asia extension of the Human Capital Index (ECA-HCI) Lower Upper Subregion/country ECA-HCI bound bound Countries within the uncertainty interval Central Asia 0.391 0.380 0.403 Kazakhstan 0.424 0.416 0.438 TUR Kyrgyz Republic 0.398 0.390 0.405 ALB, GEO, MDA Tajikistan 0.330 0.314 0.343 Uzbekistan 0.389 0.377 0.401 ARM, AZE, KGZ, MDA, ROU Central Europe and Baltic countries 0.486 0.477 0.499 Bulgaria 0.411 0.402 0.422 GEO, MNE Croatia 0.469 0.461 0.478 HUN, MLT Czech Republic 0.511 0.502 0.523 ESP, LVA, PRT Estonia 0.559 0.550 0.574 CHE, CYP, DNK, FIN Hungary 0.461 0.454 0.471 HRV, MLT, SVK Latvia 0.506 0.495 0.525 CZE, DEU, ESP, PRT, RUS AUT, BEL, CHE, FIN, FRA, ISL, NOR, 0.538 0.529 0.557 Lithuania POL, SVN Poland 0.546 0.537 0.560 BEL, CHE, EST, FIN, ISL, LTU, NOR Romania 0.387 0.377 0.400 ARM, AZE, KGZ, MDA, UZB Slovak Republic 0.458 0.451 0.468 HUN Slovenia 0.537 0.529 0.548 AUT, BEL, FRA, ISL, LTU, NOR, POL Eastern Europe 0.477 0.461 0.498 CZE, DEU, ESP, GRC, ITA, LUX, LVA, 0.492 0.477 0.512 Belarus UKR Moldova 0.391 0.381 0.408 ALB, ARM, GEO, KGZ, ROU, UZB Ukraine 0.480 0.464 0.503 BLR, DEU, GRC, HRV, ITA, LUX, MLT Northern Europe 0.564 0.553 0.579 Denmark 0.562 0.551 0.577 CHE, CYP, EST, FIN Finland 0.552 0.542 0.566 BEL, CHE, CYP, DNK, EST, NOR, POL AUT, BEL, CHE, DNK, EST, FIN, FRA, 0.541 0.527 0.562 Iceland LTU, NOR, POL, SVN BEL, CHE, EST, FIN, FRA, ISL, LTU, POL, 0.545 0.535 0.561 Norway SVN Sweden 0.583 0.571 0.597 GBR, IRL, NLD AUT, BEL, CHE, CZE, ESP, FIN, FRA, ISL, 0.525 0.506 0.554 Russian Federation LTU, LVA, NOR, POL, PRT, SVN South Caucasus 0.386 0.375 0.401 Armenia 0.385 0.376 0.398 AZE, MDA, ROU, UZB Azerbaijan 0.381 0.368 0.396 ARM, MDA, ROU, UZB Georgia 0.402 0.392 0.417 ALB, BGR, KGZ, MNE Southern Europe 0.494 0.487 0.504 Cyprus 0.564 0.548 0.581 CHE, DNK, EST, FIN, GBR, NLD Greece 0.488 0.479 0.501 BLR, DEU, ITA, LUX, UKR Italy 0.480 0.473 0.489 GRC, LUX, UKR Malta 0.470 0.462 0.482 HRV, ITA, UKR 30 Portugal 0.520 0.512 0.530 RUS Spain 0.507 0.501 0.517 CZE, DEU, LVA Turkey 0.426 0.420 0.435 KAZ Western Balkans 0.406 0.395 0.419 Albania 0.401 0.393 0.414 BGR, GEO, KGZ Bosnia and Herzegovina 0.362 0.352 0.373 MKD Kosovo Montenegro 0.415 0.396 0.428 ALB, BGR, GEO, KAZ, KGZ, TUR North Macedonia 0.359 0.353 0.368 BIH Serbia 0.443 0.429 0.457 Western Europe 0.541 0.530 0.556 BEL, FRA, ISL, LTU, NOR, POL, RUS, 0.533 0.522 0.549 Austria SVN CHE, CYP, DNK, EST, FIN, FRA, ISL, 0.548 0.536 0.566 Belgium LTU, NOR, POL, SVN AUT, BEL, CHE, FIN, ISL, LTU, NOR, 0.537 0.526 0.555 France POL, SVN Germany 0.501 0.491 0.512 BLR, CZE, ESP, LVA Ireland 0.587 0.575 0.607 GBR, NLD, SWE Luxembourg 0.487 0.475 0.503 BLR, DEU, GRC, ITA, UKR Netherlands 0.581 0.568 0.597 GBR, IRL, SWE Switzerland 0.553 0.542 0.570 BEL, CYP, DNK, EST, FIN, NOR, POL United Kingdom 0.580 0.570 0.596 IRL, NLD, SWE Simple average 0.481 0.471 0.496 Population-weighted average 0.496 0.484 0.511 Source: Authors’ calculations. 6. Concluding Remarks This paper provides an extension of the Human Capital Index that makes it more relevant for the education and health challenges faced by countries in Europe and Central Asia. Specifically, the extension incorporates two elements that are particularly important for the region. First, there is an additional focus on quality adjusted years of tertiary education, in addition to basic education. Second, health status is captured by including risk factors such as obesity, smoking and heavy alcohol consumption, all of which are prevalent in the region. This exercise highlights the importance of investing in tertiary education for many countries in the region, as well as the importance of preventing risk factors for noncommunicable and infectious diseases in the aging societies of the region. As in any cross-country benchmarking exercise, there are limitations. When analyzing the contribution to productivity from higher education, the ECA-HCI does not distinguish between types of disciplines and the 31 measure of quality can be imprecise. Moreover, data on tertiary attainment are missing for some countries. In terms of the health component, the contribution of adult health risk factors to productivity is based on estimates from the literature which can be imprecise. In any case, the ECA-HCI is not to be interpreted as a measure of welfare but as a reference for policy makers on the productivity gains that can be expected from investing in the different aspects of human capital in Europe and Central Asia. Despite these caveats, the extension of the Human Capital Index presented in this paper could be useful for all middle-income countries where investments in improving tertiary education and limiting health risk factors are likely to be priorities. 32 References Andrews, R.; Li, J. and M. Lovenheim (2016) “Quantile Treatment Effects of College Quality on Earnings” in Journal of Human Resources, vol. 51, n.1: 201-238 Brewer, D.; Eide, E. and R. Ehrenberg (1999) “Does It Pay to Attend an Elite Private College? Cross- Cohort Evidence of the Effects of College Type on Earnings” in Journal of Human Resources, vol. 34, n. 1: 104-123 Chetty, R.; Friedman, J.; Saez, E.; Turner, N. and D. Yaggan (2017) “Mobility Report Cards: The Role of Colleges in Intergenerational Mobility” NBER Working Paper 23618 Dale, S. and A. Krueger (2002) “Estimating the Payoff to Attending a More Selective College: an Application of Selection on Observables and Unobservables” in Quarterly Journal of Economics, vol. 117, n.4: 1491-1528 Dale, S. and A. Krueger (2014) “Estimating the Effects of College Characteristics over the Career Using Administrative Earnings Data” in Journal of Human Resources, vol. 49, n.2: 324-350 Eurostat (2018) Quality report of the second wave of the European Health Interview survey. Eurostat statistical reports, Publications office of the European Union: Luxembourg. Kraay, A. 2019. “The World Bank Human Capital Index: A Guide.” World Bank Research Observer 34 (1): 1–33. Loyalka, P. et al. 2019. “Computer Science Skills Across China, India, Russia and the United States.” Proceedings of the National Academy of Sciences 116 (4): 6732-6736 Smith, O. and S.N. Nguyen (2013) Getting Better: Improving Health System Outcomes in Europe and Central Asia, Europe and Central Asia Reports, World Bank: Washington, DC Wu, Z., and J. McGoogan. 2020. “Characteristics of and Important Lessons from the Coronavirus Disease 2019 (COVID-19) Outbreak in China” JAMA. Published online February 24, 2020 Zhou, F., and others. 2020. “Clinical Course and Risk Factors for Mortality of Adult Inpatients with COVID-19 in Wuhan, China: A Retrospective Cohort Study.” The Lancet 395: 1054–62. 33 Table A.1 – Productivity effect of university quality Panel a Dependent variable: log annual earnings in 2014 Ranking THE (Overall) THE (RTC) QS (Overall) QS (RTC) Both Males Females Both Males Females Both Males Females Both Males Females (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Quality score (0-100) 0.0032*** 0.0039*** 0.0026*** 0.0031*** 0.0039*** 0.0026*** 0.0027*** 0.0033*** 0.0021*** 0.0024*** 0.0029*** 0.0018*** (0.0006) (0.0007) (0.0006) (0.0006) (0.0007) (0.0005) (0.0005) (0.0005) (0.0006) (0.0005) (0.0005) (0.0006) Log parental earnings 0.3222*** 0.3526*** 0.2650*** 0.3194*** 0.3489*** 0.2628*** 0.3035*** 0.3327*** 0.2466*** 0.3076*** 0.3365*** 0.2507*** (0.0221) (0.0246) (0.0216) (0.0221) (0.0247) (0.0215) (0.0235) (0.0262) (0.0235) (0.0242) (0.0271) (0.0245) Age 0.1097*** 0.1242*** 0.0919*** 0.1097*** 0.1242*** 0.0919*** 0.1120*** 0.1270*** 0.0934*** 0.1121*** 0.1271*** 0.0935*** (0.0018) (0.0019) (0.0017) (0.0018) (0.0018) (0.0017) (0.0018) (0.0020) (0.0019) (0.0019) (0.0020) (0.0019) % of STEM majors in 0.0056*** 0.0053*** 0.0046*** 0.0057*** 0.0054*** 0.0046*** 0.0057*** 0.0051*** 0.0047*** 0.0058*** 0.0053*** 0.0049*** college (0-100) (0.0005) (0.0005) (0.0007) (0.0005) (0.0005) (0.0007) (0.0005) (0.0005) (0.0007) (0.0005) (0.0005) (0.0007) Constant 3.3955*** 2.6932*** 4.5220*** 3.4290*** 2.7378*** 4.5480*** 3.6132*** 2.9394*** 4.7470*** 3.5621*** 2.8922*** 4.6973*** (0.2640) (0.2862) (0.2509) (0.2642) (0.2972) (0.2499) (0.2910) (0.3235) (0.2896) (0.3005) (0.3342) (0.3019) Observations 1,823 1,823 1,816 1,823 1,823 1,816 1,708 1,696 1,705 1,708 1,696 1,705 Number of colleges 154 154 154 154 154 154 145 144 145 145 144 145 Panel b Dependent variable: log annual earnings in 2014 Ranking ARWU (RTC) CWUR (Overall) U-Multirank (overall) U-Multirank (RTC) Both Males Females Both Males Females Both Males Females Both Males Females (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Quality score (0-100) 0.0045*** 0.0056*** 0.0035*** 0.0073*** 0.0102*** 0.0045*** 0.0040*** 0.0047*** 0.0031*** 0.0032*** 0.0035*** 0.0028*** (0.0007) (0.0008) (0.0008) (0.0013) (0.0013) (0.0013) (0.0007) (0.0008) (0.0008) (0.0006) (0.0006) (0.0007) Log parental earnings 0.3255*** 0.3546*** 0.2730*** 0.3255*** 0.3190*** 0.2641*** 0.3204*** 0.3498*** 0.2675*** 0.3131*** 0.3421*** 0.2605*** (0.0187) (0.0216) (0.0183) (0.0187) (0.0147) (0.0143) (0.0196) (0.0230) (0.0171) (0.0202) (0.0239) (0.0172) Age 0.1105*** 0.1252*** 0.0921*** 0.1078*** 0.1240*** 0.0894*** 0.1141*** 0.1297*** 0.0951*** 0.1139*** 0.1295*** 0.0949*** (0.0019) (0.0021) (0.0019) (0.0013) (0.0015) (0.0013) (0.0019) (0.0021) (0.0018) (0.0019) (0.0021) (0.0018) % of STEM majors in 0.0059*** 0.0054*** 0.0053*** 0.0057*** 0.0052*** 0.0047*** 0.0055*** 0.0053*** 0.0040*** 0.0060*** 0.0059*** 0.0043*** college (0-100) (0.0006) (0.0005) (0.0007) (0.0005) (0.0005) (0.0006) (0.0008) (0.0008) (0.0009) (0.0008) (0.0008) (0.0009) Constant 3.3484*** 2.6820*** 4.4156*** 3.2907*** 2.5193*** 4.3775*** 3.1628*** 2.4354*** 4.3134*** 3.2624*** 2.5575*** 4.3854*** (0.2291) (0.2655) (0.2207) (0.1774) (0.2655) (0.1701) (0.2649) (0.3107) (0.2224) (0.2690) (0.3193) (0.2204) Observations 1,869 1,868 1,865 3,302 3,252 3,278 2,006 1,972 1,985 2,006 1,972 1,985 34 Number of colleges 158 158 158 279 275 278 170 167 169 170 167 169 Note: The common sample is composed of universities which are present in all the six rankings. Clustered standard errors at the college level in parentheses. Significance: * p<0.10, ** p<0.05, *** p<0.01. Panel c Dependent variable: log annual earnings in 2014 Ranking US News (overall) Aggregate Quality Score Agg. Q. Score (common sample) Both Males Females Both Males Females Both Males Females (1) (2) (3) (4) (5) (6) (7) (8) (9) Quality score (0-100) 0.0019*** 0.0027*** 0.0013** 0.0024*** 0.0031*** 0.0016*** 0.0044*** 0.0052*** 0.0036*** (0.0006) (0.0006) (0.0006) (0.004) (0.0004) (0.0004) (0.0009) (0.0010) (0.0008) Log parental earnings 0.3440*** 0.3701*** 0.2934*** 0.2986*** 0.3142*** 0.2646*** 0.3202*** 0.3597*** 0.2543*** (0.0206) (0.0236) (0.0202) (0.0136) (0.0150) (0.0134) (0.0248) (0.0283) (0.0225) Age 0.1074*** 0.1228*** 0.0891*** 0.1074*** 0.1237*** 0.0894*** 0.1157*** 0.1295*** 0.0979*** (0.0016) (0.0018) (0.0016) (0.0013) (0.0014) (0.0013) (0.0023) (0.0024) (0.0022) % of STEM majors in 0.0059*** 0.0059*** 0.0050*** 0.0058*** 0.0053*** 0.0049*** 0.0046*** 0.0043*** 0.0032*** college (0-100) (0.0004) (0.0004) (0.0006) (0.0005) (0.0005) (0.0006) (0.0008) (0.0008) (0.0010) Constant 3.2374*** 2.565*** 4.2977*** 3.8250*** 3.2798*** 4.6636*** 3.2758*** 2.4979*** 4.5139*** (0.2415) (0.28001 (0.2293) (0.1679) (0.1881) (0.1606) (0.3061) (0.3469) (0.2707) Observations 2,363 2,363 2,360 3,784 3,689 3,738 1,159 1,159 1,156 Number of colleges 199 199 199 323 315 321 98 98 98 Note: The common sample is composed of universities which are present in all the six rankings. Clustered standard errors at the college level in parentheses. Significance: * p<0.10, ** p<0.05, *** p<0.01. 35 Appendix A. Skill-based adjustment of higher education Quality adjustment of higher education can be performed by measuring the quality of inputs (educational institutions) or the quality of outputs (academic proficiency of graduates from higher education). Quality adjustment using university rankings corresponds to the former approach. Quality adjustment using the skills of university graduates corresponds to the latter approach. Adult skill proficiency is multidimensional. This analysis focuses on two dimensions that are measured by the Program for the International Assessment of Adult Competencies (PIAAC) survey: literacy proficiency and numeracy proficiency. The PIAAC survey, run by the Organisation for Economic Co-operation and Development, has been carried out in 40 countries, of which 24 are in Europe and Central Asia. The Skills Towards Employment survey, which is run by the World Bank, measures literacy proficiency on a scale equivalent to the PIAAC in three additional countries in Europe and Central Asia. The literacy and numeracy proficiencies are measured on a 0–500 scale; any value greater than 376 is considered highly proficient. The benchmark for full proficiency is set at 400, which exceeds the value reported at the 90th percentile of the score distribution of the average adult population in all countries. Each skill type is weighted equally. The quality-adjustment measure used is the proficiency in both types of skills of individuals 30–34 who completed a tertiary degree in each country. This demographic group was chosen to match the group for which attainment rates of tertiary degrees are used. The skill-adjusted years of higher education (SAYH) is then derived using the following formula: 30−34 30−34 30−34 1 = × 3.5 × � + � 400 400 2 The correlation between the SAYH and the QAYH is very high. It is similar for literacy (figure A.1, panel a) and numeracy (figure A.1, panel b) skills. Adjusting the quality of higher education based on adult skill proficiency (SAYH) or university rankings (QAYH) seems to yield similar results. 36 Figure A.1 Correlation between skill-adjusted and quality-adjusted years of higher education for literacy and numeracy skills Source: Authors’ calculations. 37 This finding is not surprising, given that there is a high correlation between adult skill proficiency and the average score of a country’s universities in the six university rankings (figure A.2). Figure A.2 Correlation between skill proficiency and university ranking quality score for literacy and numeracy skills Note: The standardized quality score for higher education is calculated in the following way: The quality scores from each of the six university rankings (the Times Higher Education, the Quacquarelly Symonds, the Academic Ranking of World Universities, the Center for World University Rankings, the U.S. News U.S. Global Universities Ranking, and U-Multirank) are first standardized to a global mean of 0 and a standard deviation of 1 and then averaged for every country. This value is then rescaled to range from 0 to 100 for presentational purposes. PIAAC = Programme for the International Assessment of Adult Competencies. 38 Appendix B. Estimates of the Effect of Adult Health Risk Factors on Productivity This appendix reports conditional estimates on log earnings. The characteristics controlled for may differ across papers, but they always include age, gender, and education. Table B.1 Review of studies on effect of obesity on productivity Estimate Source Paper Low High Average Comment in paper Averett and –0.03 –0.15 –0.09 Coefficients compare obese people (BMI > 30) Table 4 Korenman and people of ideal weight (BMI 20–25). Low (1996) estimate is for men, 1988 sample; high estimate is for women, 1981 sample. Cawley, Grabka, 0 –0.1986 –0.0993 Coefficients compare obese people (BMI > 30) Table 2 and Lillard and people of ideal weight (BMI 20–25). Low (2005) estimate is for men in the United States (not significantly different from zero); high estimate is for women in the United States. Lundborg and –0.058 –0.074 –0.066 Coefficients compare obese people (BMI > 30) Table 9 others (2007) and non-obese people (BMI < 30); high estimate includes health status as control. Brunello and –0.04 –0.105 –0.0725 Regression is linear specification with BMI as Table 3 D’Hombres independent variable. Coefficients are (2007) multiplied by 5 to simulate a change from BMI 25 to BMI 30. Low estimate is for women, controlling for occupation and sector; high estimate is for men, not controlling for occupation and sector. Kline and Tobias –0.0685 –0.153 –0.1108 Regression is nonlinear specification with BMI Table (2008) as independent variable. Low estimate IV corresponds to expected change between BMI 25 and BMI 30 for women; high estimate corresponds to same change for men. Lundborg, –0.072 –0.153 –0.1125 Coefficients compare obese people (BMI > 30) Table Nysted, and and people of ideal weight (BMI 20–25). Low 4.1, Rooth (2010) estimate is for specification controlling for columns noncognitive skills; high estimate is for C, D, E specification not controlling for any skill. Bockerman and 0 –0.355 –0.1775 Regression is linear specification with BMI as Table 1 others (2019) independent variable. Coefficients are multiplied by 5 to simulate a change from BMI 25 to BMI 30. Low estimate corresponds to genetic instrumental variable 97 SNP (not significantly different from zero). High 39 estimate corresponds to genetic instrumental variable 32 SNP. Median –0.0993 J. Viinikainen, T. Lehtimäki, S. Rovio, I. Seppälä, J. Pejkonen, and O. Raitakari. 2019. “The Effect of Weight on Labor Market Outcomes: An Application of Genetic Instrumental Variables.” Health Economics 28: 65–77. Brunello, G., and B. D’Hombres. 2007. “Does Body Weight Affect Wages? Evidence from Europe.” Economics & Human Biology 5 (1): 1–19. Cawley, J., M. Grabka, and D. Lillard. 2005. "A Comparison of the Relationship between Obesity and Earnings in the U.S. and Germany." Schmollers Jahrbuch: Journal of Applied Social Science Studies / Zeitschrift für Wirtschafts–und Sozialwissenschaften 125 (1): 119–29. Kline, B., and J. Tobias. 2008. “The Wages of BMI: Bayesian Analysis of a Skewed Treatment–Response Model with Nonparametric Endogeneity.” Journal of Applied Econometrics 23: 767–93. Lundborg, P., K. Bolin, S. Hojgard, and B. Lindgren. 2007. “Obesity and Occupational Attainment Among the 50+ of Europe.” Advances on Health Economics and Health Services Research 17: 219–51. Lundborg, P., P. Nysted, and D.–O. Rooth. 2010. “No Country for Fat Men? Obesity, Earnings, Skills and Health among 450,000 Swedish Men.” IZA Discussion Paper No. 4775, Institute of Labor Economics, Bonn. 40 Table B.2 Review of studies on effect of smoking on productivity Estimate Source Paper Low High Average Comments in paper Levine, –0.04 –0.08 –0.06 Coefficients compare smokers (more than 1 Table 4 Gustafson, and cigarette a day) and nonsmokers. Low estimate Velenchik (1997) is for 1984; high estimate is for 1991. Van Ours (2004) –0.085 –0.119 –0.102 Coefficients compare smokers and Table nonsmokers. Low estimate is for average 10 smokers; high estimate is for twice average smokers. Auld (2005) –0.083 –0.268 –0.1755 Coefficients compare smokers and Table 2 nonsmokers. Low estimate treats smoking as exogenous; high estimate treats smoking as endogenous. Grafova and –0.076 –0.102 –0.089 Coefficient compare persistent smokers and Table 7 Stafford (2009) people who never smoked. Low estimate is for 1986; high estimate is for 2001. Lokshin and –0.19 –0.23 –0.21 Coefficient corresponds to (causal) difference Table 2 Beegle (2011) in earnings of current smokers and and nonsmokers. Low estimate is for LIV page specification; high estimate is for 2SLS 227 specification. Bondzie (2016) –0.043 –0.069 –0.056 Matching estimates of differences between Table 5 smokers and nonsmokers. Low estimate corresponds to kernel ATT; high estimate corresponds to nearest neighbor ATT. Median –0.096 References for Table B.2 Auld, C. 2005. “Smoking, Drinking and Income.” Journal of Human Resources 40 (2): 505–18. Bondzie, E. A. 2016. “Effect of Smoking and Other Economic Variables on Wages in the Euro Area.” MPRA Paper No. 69230, University of Munich, Germany. Grafova, I., and F. P. Stafford. 2009. “The Wage Effects of Personal Smoking History.” Industrial and Labor Relations Review 62 (3): 381–93. Levine, P. B., T. A. Gustafson, and A. D. Velenchik. 1997. "More Bad News for Smokers? The Effect of Cigarette Smoking on Wages." Industrial and Labor Relations Review 50 (3): 493–509. Lokshin, M., and K. Beegle. 2011. “Foregone Earnings from Smoking: Evidence for a Developing Country.” Research in Labor Economics 33: 209–38. 41 Van Ours, J. 2004. “A Pint a Day Raises a Man’s Pay; but Smoking Blows That Gain Away.” Journal of Health Economics 23 (5): 863–86. 42 Table B.3 Review of studies on effect of heavy drinking on productivity Estimate Source Paper Low High Average Comments in paper Mullahy and –0.163 –0.176 –0.1695 Coefficients compare people diagnosed with Table 3, Sindelar (1993) alcoholism and people not diagnosed with all obs. alcoholism. Low estimate is for people ever diagnosed with alcoholism; high estimate is for people diagnosed with alcoholism in past year. Hamilton and –0.254 –0.758 –0.506 Coefficients correspond to decomposition of Table 4 Hamilton (1997) wage differences attributed to differences in and returns to characteristics of heavy drinkers page (people who consume eight or more drinks on 148 one or more days in the previous week) and nondrinkers. Low estimate is for wider definition of heavy drinker. Zarkin and 0.082 –0.021 0.0305 Coefficients compare heavy drinkers (people Table 2 others (1998) who consumed more than 94 drinks in past 30 days for men, 48 drinks for women) and nondrinkers. Low estimate is for men; high estimate is for women. Barrett (2002) –0.08 –0.19 –0.135 Low estimate compares heavy drinkers (people Table 4 who consumed eight or more drinks on one or more days the previous week) and nondrinkers. High estimate is for heavy drinkers versus moderate drinkers. Sloan and 0 –0.459 –0.2295 Coefficient compares heavy drinkers (people Table 2 Grossman (2011) who consume more than 12 drinks a week) and nondrinkers. Low estimate is for whites and women (not significantly different from zero); high estimate is for black men. Bockerman, –0.18 –0.424 –0.302 Coefficient corresponds compares heavy Table V Hyytinen, and drinkers (men who consume more than 280 Maczulskij grams of alcohol a week and women who (2017) consume more than 190) and moderate drinkers (men who consume less than 280 grams of alcohol a week and women who consume less than 190). Low estimate is for twin differences in monozygotic twins; high estimate is for twin differences in dizygotic twins. Median –0.1995 References for Table B.3 43 Barrett, G. 2002. “The Effect of Alcohol Consumption on Earnings.” Economic Record 78 (1): 79–96. Bockerman, P., A. Hyytinen, and T. Maczulskij. 2017. “Alcohol Consumption and Long–Term Labor Market Outcomes.” Health Economics 26: 275–91. Hamilton, V., and B. H. Hamilton. 1997. “Alcohol and Earnings: Does Drinking Yield a Wage Premium?” Canadian Journal of Economics 30 (1): 135–51. Mullahy, J., and J. L. Sindelar. 1993. “Alcoholism, Work and Income.” Journal of Labor Economics 11: 494–519. Sloan, F. A., and D. S. Grossman. 2011. “Alcohol Consumption in Early Adulthood and Schooling Completed and Labor Market Outcomes at Midlife by Race and Gender.” American Journal of Public Health 101 (11): 2093–2101. Zarkin, G. A., M. T. French, T. Mroz, and J. W. Bray. 1998. “Alcohol Use and Wages: New Results from the National Household Survey on Drug Abuse.” Journal of Health Economics 17: 53–68. 44 45