|
on Education |
|
Issue of 2026–10–19
sixteen papers chosen by João Carlos Correia Leitão, Universidade da Beira Interior |
| By: | Shruti Jha; John A. List; Rebecca C. Royer; Anya Samek |
| Abstract: | Do early investments in parents pay off even when treatment effects for children appear to fade out quickly? We answer this question by leveraging a field experiment in which households with 3-5 year-old children were randomized into a parenting program, a preschool, or a control group, combined with a natural experiment (Covid-19 school closures) that occurred ten years later. Prior to the closures, treatment effects on academic achievement had faded to zero across all groups. Yet, with Covid-19 school closures, effects of the parenting program re-emerge: test scores in control and preschool groups drop by roughly 0.25-0.30 SD, whereas scores in the parenting group drop by only 0.11 SD (difference: p = 0.01). Our findings suggest that parenting programs can generate latent human capital that only becomes visible when complementary school inputs are removed. This result is consonant with parental and school investments acting as substitutes in skill production. Because this latent parental capacity pays off precisely when school inputs fail, standard measurement tools for early childhood programs can miss crucial state-contingent value. |
| JEL: | C93 I26 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35851 |
| By: | Elacqua, Gregory; Munevar, Isabela; Prada, María Fernanda; Rodríguez, Julio |
| Abstract: | This paper analyzes the association between the composition of school spending and various measures of school effectiveness. Using audited financial accounts for the universe of publicly funded urban schools in Chile between 2015 and 2022, disaggregated into six expenditure categories and 37 subcategories, we link spending patterns to school value-added measures of test scores, attendance, chronic absenteeism, and dropout. Aggregate spending per student is only weakly associated with school effectiveness. In contrast, spending composition matters substantially: we reject equal associations across expenditure categories for all outcomes, and no category is associated with all four dimensions of effectiveness. Personnel spending is associated with higher test-score value added but not attendance, whereas operations spending is associated with higher attendance and lower chronic absenteeism but not with test score gains. Infrastructure spending is associated with both learning and attendance outcomes. We also find considerable heterogeneity across school types and student populations. The categories prioritized by Chiles targeted equity subsidy, quality programs and student well-being, are associated with higher test-score value added primarily in the most disadvantaged schools, while operations and infrastructure expenditures exhibit opposite associations in public and private voucher schools. The estimated associations are modest in magnitude: a one-standard-deviation increase in the most predictive spending categories corresponds to 2 to 7 percent of the crossschool dispersion in value added. Nevertheless, the findings have important policy implications because education funding policies typically target specific expenditure categories rather than overall spending levels. Overall, the results suggest that the composition of spending is more informative than expenditure levels alone and that evaluating education finance policies using a single outcome or pooled estimates can understate the benefits of targeted investments. |
| Keywords: | School Spending Composition;School Value Added |
| JEL: | I22 I25 H75 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:idb:brikps:14772 |
| By: | David Card; Laura Giuliano |
| Abstract: | We study an accelerated 3-year math curriculum offered in the middle schools of a large urban district that enables students to finish Algebra I and Geometry before high school. Entry to the program is based on 5th-grade math test scores, with a cutoff at the 80th percentile. Continuing eligibility requires at least a B grade in the previous year; successful completion of all three years is highly correlated with an index of non-cognitive skills. Following students for 6 years after high school graduation, we find that the impacts of admission vary widely by non-cognitive skill and gender. Marginally eligible boys with high non-cognitive skills have a 33 percentage point (ppt.) increase in the probability of completing calculus by 11th grade and a 12 ppt. increase in the probability of completing a major in STEM or business within 6 years; boys with lower noncognitive skills take more advanced math classes in high school but have no significant effects on their college outcomes. Like their male peers, marginally eligible girls with high non-cognitive skills are more likely to complete calculus by 11th grade (+17 ppt.). Their rate of graduating with a STEM or business major is unaffected, but they switch from social sciences to the core STEM majors (science, engineering and math). Girls in the two lower terciles of non-cognitive skill are also strongly affected by entry to the program, with higher participation in advanced math classes in high school, higher rates of entry to selective 4-year colleges, and higher rates of completing a STEM or business major (+12 ppt). |
| JEL: | I21 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35847 |
| By: | Ajzenman, Nicolás; Del Toro Mijares, Ana Teresa; Elacqua, Gregory; Hermosilla, Catalina; Veleda, Santiago Nicolas |
| Abstract: | A central obstacle to scaling many interventions is implementer quality: as a program expands and the pool of implementers broadens, average quality falls. Conversation-based interventions (counseling, mentoring) are particularly sensitive to this constraint. LLMs offer a natural way out: personalized exchanges at zero marginal cost and uniform quality. We run a nationwide experiment in Chile (N42, 000) designed to reduce teacher shortages by increasing applications to education majors. We compare an AI chatbot, Kai, with trained human counselors, both reaching high school seniors via WhatsApp. Among students with baseline interest in education (the interventions target), Kai significantly increases first-ranked education majors and the share of education majors in students choice sets; estimates in the human arm are smaller and not significant. Although the best human counselors are highly effective, Kai matches or outperforms roughly two- thirds of them. Its advantage lies in what we call “controlled variance”: it compresses the dispersion in quality observed across human implementers. Finally, text analysis reveals different interaction patterns across arms. Kai concentrates on factual content and scores higher on semantic coherence, while counselors score substantially higher on motivational and empathetic language. |
| Keywords: | Teachers;teacher shortages;Sclae-up;behavioral interventions;AI chatbots;Generative AI |
| JEL: | D91 I23 I25 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:idb:brikps:14771 |
| By: | Claudia Allende; Francisco Gallego; Christopher Neilson |
| Abstract: | We study how information about school quality affects families’ choices and how scaling up the intervention changes its impact. In a randomized trial in Chile, we offer families a short video and a personalized school report card before they enroll their children. Treated families choose schools with higher test scores and value added. Among families reached before enrollment, treated children score about 0.2 standard deviations higher in fourth grade, with positive but less precise estimates on later assessments. We embed the experimentally identified demand shift in a structural model of school choice and competition to study equilibrium effects at scale. Capacity constraints and centralized assignment attenuate the demand-side gains, but the same shift in demand strengthens schools’ incentives to improve quality, and the supply response roughly offsets the congestion effect. More generally, equilibrium scaling can alter experimental effects in either direction because congestion and endogenous quality responses work against each other. We use the model to guide the design of a follow-up cluster-randomized experiment, focusing on the tradeoff between precision and contamination from equilibrium spillovers. |
| JEL: | L10 L15 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35837 |
| By: | Canh Thien Dang; An Nguyen |
| Abstract: | Experimental studies of generative AI in education mostly report average effects, yet field evidence shows that AI can narrow attainment gaps or widen them. We argue that the direction depends on the judgement burden, the expertise a learner must supply to screen AI output before learning from it, and that expert verification before release moves this burden from students to an accountable tutor. We test the argument in a two-cohort difference-in-differences design in which one half of a compulsory firstyear university economics course received AI-generated podcasts, FAQs and quiz-based study guides, produced with a source-grounded model and checked by a named graduate teaching assistant (170 students; 340 examination marks). Access was associated with a 2.34-mark advantage on a 50-mark component. The share of marks below the upper-second classification boundary fell by 24.7 percentage points relative to the counterfactual, effects were significant at every threshold from 23 to 31 marks and at none above, and roughly three-quarters of the average originated in the bottom quintile. The threshold estimate is robust to removing the lowest-scoring students from the pre-intervention cohort; the average effect is not. Interviews and feedback from 36 students indicate that the verification label gave students a reason to engage with AI-generated material without ending their scrutiny of it. Evaluations of AI learning resources that report only mean effects cannot detect whether the students the resources are meant to help are the ones who gain. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.07097 |
| By: | Pavese, Caterina (Università di Bologna, CESifo); Meulen, Simon ter (ifo Institute, University of Munich, CESifo); Woessmann, Ludger (University of Munich, ifo Institute; Hoover Institution, Stanford University; CESifo, IZA@LISER, and RFBerlin) |
| Abstract: | Does direct integration into native schools or temporary placement in separate schools better promote the education and cultural assimilation of refugee children? In the Netherlands, primary school-aged refugee children are placed in either integrated or separated schools depending on their randomly assigned municipality. By contrast, all secondary-school-aged arrivals are initially separated. We exploit this institutional setting in two-way fixed-effects models with age at-arrival and municipality fixed effects. Using rich administrative data, we find that integrated schooling reduces educational attainment at age 18, particularly for children of lower-educated parents. At the same time, integrated schooling increases cultural assimilation, measured by hormonal-contraceptive usage among teenage women, particularly for children from non Muslim-majority countries. A distinctive feature of separated schooling that may account for its positive educational effect is that it allows employing teachers specialized in native-language teaching for non-native speakers. |
| Keywords: | refugee children, schools, integration, educational attainment, contraceptives JEL Classification: I24, I21, J13, J15, O15 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:cge:wacage:826 |
| By: | Cristóbal Castro Barrientos (Auckland University of Technology) |
| Keywords: | higher education, returns to education, qualifications, vocational education, gender, New Zealand |
| JEL: | I26 J31 J24 J16 |
| Date: | 2026–10–05 |
| URL: | https://d.repec.org/n?u=RePEc:aut:wpaper:2026-04 |
| By: | Claus Thustrup Kreiner; Paul Bingley; Benjamin Ly Serena |
| Abstract: | Socioeconomic inequality in longevity is typically measured using a single indicator, such as education or income. When indicators are imperfect proxies for socioeconomic status, this approach understates inequality. Using multiple indicators can improve measurement by better capturing latent socioeconomic status. We combine education, income, occupation, wealth, and IQ scores using machine learning. Using Danish population-wide data spanning 40 years, we estimate life expectancy at age 40 for the 1942-44 birth cohorts. Individuals at the top of the socioeconomic distribution live nearly 25 years longer than those at the bottom. The estimated gradient is 50-150% steeper with multiple indicators. |
| Keywords: | Life Expectancy, Inequality, Machine Learning |
| JEL: | I14 |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:crm:wpaper:26234 |
| By: | Daniel Aaronson; Mark Borgschulte; Sunny Liu; Bhashkar Mazumder |
| Abstract: | Does education lead to political activity? We find that a large, early 20th century school construction program, the Rosenwald schools, led to more NAACP branches in the rural South starting in the early 1940s, i.e. prior to the classic U.S. Civil Rights Movement. Branch growth was stronger where Jim Crow occupational segregation prevented Black men from realizing some of the economic gains of their education. Our findings lend support to theories emphasizing that economic context can act as a critical mediator in determining whether human capital causes political engagement. |
| JEL: | I20 J71 N32 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35842 |
| By: | Liu Zhang; Mark Esposito |
| Abstract: | Firms increasingly evaluate deployed AI agents using repeated human ratings, yet observed score changes may reflect the measurement process as much as changes in the agent itself. Firms also rarely know how long an evaluation remains informative as deployed systems evolve. We study reliability as a latent state using 2, 611 rubric-scored production interviews from a deployed voice-and-video interviewing agent, overlapping human reviewers, the agent's change log, and support tickets. Reviewer severity varied substantially: two reviewers assessing the same batches differed by 0.79 standard deviations on the composite index, and a shift in reviewer composition flattened the raw trend. After adjusting for reviewer effects and smoothing batch-level estimates, evaluated reliability increased by 0.53 standard deviations from March to August 2026. Reliability also shifted materially across logged deploys. At the estimated rate of unaccounted movement, forecast uncertainty reaches the magnitude of the largest observed deploy contrast after about five weeks, although this horizon is imprecisely estimated and based on a small number of step-like changes. An independent operational outcome moved in the same direction: interview-related support tickets fell by about 10% per month relative to technical-issue tickets. These data do not identify a causal effect of evaluation on performance. Instead, they show that longitudinal AI evaluation requires three checks: comparability across raters, evidence currency under drift, and corroboration with consequential operational outcomes. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.07003 |
| By: | Leandro Carvalho; Sjoerd van Alten; Silvia Barcellos; Stephen Dorn; Titus Galama; Catharina A. Hartman; Qiongshi Lu; Dilnoza Muslimova |
| Abstract: | Personality is associated with economic outcomes, but whether these relationships are causal remains unclear. We exploit within-family variation in genetic propensity toward the Big Five generated by random genetic transmission from parents to offspring. Combining polygenic indexes from the Lifelines cohort study with 19 years of administrative tax records in the Netherlands, we find that greater genetic propensity toward emotional stability increases earnings and household wealth, while genetic propensity toward conscientiousness increases household wealth. The other three traits show no consistent pattern across outcomes and specifications. The effect of genetic propensity toward emotional stability on earnings remains positive as we progressively account for genetic propensity toward educational attainment, cognitive performance, and depression. These findings strengthen the evidence that personality causally affects earnings and wealth. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.06088 |
| By: | Juan Pablo Atal; Tomás Larroucau; Pablo Muñoz; Cristóbal Otero |
| Abstract: | Occupational licensing is a widespread quality regulation that potentially increases the quality of labor but reduces its quantity. We provide a theoretical framework to empirically quantify this trade-off in public hospitals, where both clinical quality and physician shortages are critical concerns. Using quasi-exogenous variation driven by a recent and unprecedented migration of physicians to Chile, together with growth in domestic medical schools, we show that more physicians improve patient outcomes, while lower-quality physicians—as measured by licensing exam performance—worsen them. Incorporating these findings into our theoretical framework, we evaluate the implications of locally changing the stringency of the licensing policy. |
| JEL: | I18 J24 J44 J45 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35855 |
| By: | Samuel Fiifi Eshun; Evžen Kočenda |
| Abstract: | How robust is the empirical evidence on the determinants of financial inclusion, and why do reported effects differ across studies? We provide the first systematic meta-analysis of this literature, synthesizing 5, 172 estimates from 100 studies published between 2013 and 2024. We standardize reported estimates using partial correlation coefficients corrected with the UWLS+3 procedure and examine publication selection using precision-weighted FAT-PET regressions. We employ Bayesian model averaging to identify sources of heterogeneity. Publication selection is substantial but differs markedly across determinant categories. In our main specification, significant effects beyond publication bias remain for macroeconomic conditions, banking and financial depth, technology and infrastructure, and the persistence of financial inclusion, whereas evidence for demographic, income, education and human capital, and institutional determinants is weaker. Except for persistence, the corrected associations are generally modest. Technology and infrastructure provide the most consistent evidence among actionable determinants, while the effect attributed to banking and financial depth is substantially smaller than the unadjusted literature suggests. We find no systematic evidence of p-hacking. Differences in measurement, estimation methods, sample characteristics, and economic context explain much of the remaining heterogeneity. Overall, the literature contains economically relevant regularities, but conventional estimates often overstate their magnitude and generality. |
| Keywords: | financial inclusion, banks, meta-analysis, model uncertainty, publication bias, Bayesian model averaging |
| JEL: | C83 G21 O16 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_13039 |
| By: | Levi Boxell; Jacob Conway; Nicolas Longuet-Marx |
| Abstract: | What role do journalists play in determining the political slant of the news they produce? We develop and estimate a model where journalists and newspaper outlets contract over both slant and wages. The model implies a set of conditions under which we can consistently estimate the role of journalist preferences in driving the observed variation in slant across outlets by leveraging journalist transitions between outlets. To measure slant, we train a transformer-based, machine learning model using articles tweeted by politicians and apply it to a full-text database of 9+ million newspaper articles published in the US between 2013 and 2018. Under our model and identifying assumptions, our estimates (a) reject the hypothesis that journalists have zero ideological preferences over the content they produce and (b) imply that 10% of the variation in slant can be explained by journalists. When journalists move across outlets, their average slant shifts by 77% of the gap between their destination and origin outlet averages. In counterfactuals that reweight journalist party composition, moving from the observed left-leaning distribution to an even distribution of Democrats, Republicans, and non-partisans shifts average article slant in a more conservative direction, but only by 0.04 standard deviations in slant. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.04182 |
| By: | Daniel P. Gross; Dror Shvadron; Hansen Zhang |
| Abstract: | Large language models are diffusing rapidly into science, raising concerns that delegating research writing to AI may undermine the development of expertise that scientific training is supposed to build. We measure AI-generated writing in the dissertations of a near-census of US STEM PhD graduates from 2019 to 2026 and link graduates to their subsequent employment. AI writing is absent before 2023, present in 29% of dissertations filed in 2026, and rapidly growing. Use is more common among students from non-English-speaking countries and at lower-ranked programs. Among graduates in the same program and cohort, those whose dissertations contain AI writing are less likely to enter academic research careers and more likely to enter non-tenure-track roles or industry, with this pattern strongest among US-origin students. At the same time as AI may increase research productivity, it thus appears to also be displacing a potentially important portion of the work through which scientific expertise is developed. |
| JEL: | I23 J24 O31 O33 |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:nbr:nberwo:35859 |