Measuring and forecasting progress in education: what about early childhood?

A recent Nature article modelled within-country inequalities in primary, secondary, and tertiary education and forecast progress towards Sustainable Development Goal (SDG) targets related to education (SDG 4). However, their paper entirely overlooks inequalities in achieving Target 4.2, which aims to achieve universal access to quality early childhood development, care and preschool education by 2030. This is an important omission because of the substantial brain, cognitive and socioemotional developments that occur in early life and because of increasing evidence of early-life learning’s large impacts on subsequent education and lifetime wellbeing. We provide an overview of this evidence and use new analyses to illustrate medium- and long-term implications of early learning, first by presenting associations between pre-primary programme participation and adolescent mathematics and science test scores in 73 countries and secondly, by estimating the costs of inaction (not making pre-primary programmes universal) in terms of forgone lifetime earnings in 134 countries. We find considerable losses, comparable to or greater than current governmental expenditures on all education (as percentages of GDP), particularly in low- and lower-middle-income countries. In addition to improving primary, secondary and tertiary schooling, we conclude that to attain SDG 4 and reduce inequalities in a post-COVID era, it is essential to prioritize quality early childhood care and education, including adopting policies that support families to promote early learning and their children’s education. We used data for 430,264 adolescents in 73 middle- and high-income countries surveyed in the 2018 Programme for International Student Assessments (PISA). The PISA is an international programme to assess adolescents’ reading, mathematics, and science literacy every 3 years for nationally representative samples of 15-year-old students enroled in school. The PISA also collects data on the characteristics of adolescents and their backgrounds. Among such information, the PISA asks students to retrospectively report how many years of pre-primary education they attended, following the International Standard Classification of Education Level 0 (ISCED-0). Using PISA data, we used multivariate regression models to assess the association between years of pre-primary education attendance and mathematics and science test scores. Given that PISA employs an imputation methodology to provide plausible values for each student’s test score, as a first step, we averaged such plausible values to obtain a single mathematics and science test score, which we standardized to have a mean of zero and standard deviation of one to aid interpretability. Subsequently, we estimated two models to test the association between students’ mathematics and science standardized test scores (zTestScorei) and binary variables indicating whether students attended one year (OneYeari) or two or more years (TwoYeari) of pre-primary education. We added covariates to the model in order to reduce potential bias, including adolescents’ age and gender, a wealth index provided by PISA, maternal and paternal education, and age of entry to primary school. Furthermore, we included country and subnational (geographical region or school type) fixed effects to make within-country comparisons. We estimated separate models according to the World Bank’s categorization of income groups, i.e., for lower-middle-income countries (N = 6), upper-middle-income countries (N = 24), and high-income countries (N = 43), and region, i.e., East Asia and Pacific (N = 11), Europe and Central Asia (N = 42), Latin America and the Caribbean (N = 10), Middle East and North Africa (N = 8), and North America (N = 2). To obtain the COI for not reaching the SDG 4.2 targets for 1 year, we extend a procedure used to estimate the COI related to pre-primary schooling for five Latin American countries for the Lancet series on early childhood development21,43,45,46. In Eq. (1), the increase in individual earnings in future decades as a consequence of participating in preschool (PCIj x i, where PCIj is per capita income in year j and i is the causal impact of preschool on that income) is discounted by the discount rate d and summed over the relevant years in which earnings are expected to be affected (from when the individual starts to work a years after preschool through t years of working life) and then compared with the per child programme cost (c) for the N children covered by preschool (i) for the 2018 enrolments and (ii) if the SDG 4.2 targets were attained. The COI for each country therefore depends on projections for that country’s PCI, the impact of preschool on per capita income (i), the per child programme cost (c) and the expansion in enrolment in order to obtain the SDG 4.2 targets (100%−N). Note that this procedure probably leads to conservative estimates of the COI because only effects on adult earnings are included, but other short- and long-term impacts that are hard to monetize, such as reduced crime, are omitted. On the other hand, there may be general equilibrium effects that work in the opposite direction. Information on the discount rate (d), number of children affected (N), impact of pre-primary school on adult earnings (i) and cost (c) of pre-primary is critical for the simulations. As is common for many evaluations of social programmes where benefits accrue in the long term, in our simulations we used a discount rate (d) of 3%. For the current enrolments (N) we used the 2018 gross enrolment rates (GER) reported in UNESCO, Institute for Statistics47. For the SDG-targeted enrolments we used the maximum of 100% and the actual 2018 enrolments. The latter may be over 100% because the UNESCO data estimate the total enrolments of pre-primary students of all ages to the ratio of children in the country of pre-primary school ages, and there is catch-up in some countries with older children attending pre-primary school. We assume that such catch-up is of interest in attaining the SDG 4.2 targets. This use of the UNESCO GERs causes an underestimate of the COIs. The causal evidence on long-term effects of pre-primary school from randomized experiments in which children have been subsequently followed-up in their adult years is sparse, but suggests that impacts in earnings are substantial, of the order of 14% over the lifetime48. However, since that evidence comes from high-quality small-scale interventions targeting children from low socioeconomic backgrounds, it may not be externally valid in the case of lower-quality programmes, programmes implemented at scale or with children participating from all socioeconomic backgrounds. Thus, we have adopted a lower impact value of 8% for our simulations. The estimates for per child costs of existing pre-primary services, c, vary considerably across countries49,50. Since an important part of these variations reflects differences in wages and prices for services that relate to income levels across countries, we adjusted the programme costs for each group of countries based on the price level ratio of purchasing-power-parity (PPP) conversion factors that reflect the value of wages for services better than do market exchange rates. These assumptions are strong so the COI estimates are somewhat crude for any particular country. Supplementary Table 5 provides sensitivity analyses. COI do not change substantially if the assumed impact of pre-primary school on earnings decreases one or two percentage points, or if costs increase by 10% or 20%. If we apply higher discount rates (from 4% to 5%), patterns are similar but with smaller COI. Even though estimates would probably need to be refined to provide guidance for any particular country’s polices, they provide a useful order of magnitude for understanding an important component of the long-run global economic costs of lags in reaching the SDG 4.2 targets. Further information on research design is available in the Nature Research Reporting Summary linked to this article.