For three decades, development education policy has been dominated by an implicit theory of change: get children into classrooms, and learning will follow. Billions have flowed toward school construction, fee elimination, conditional cash transfers, and universal enrollment campaigns. By the metric of access, this represents one of the great successes of development practice—primary enrollment rates in low-income countries have climbed from roughly 50 percent in 1990 to over 85 percent today.

Yet the accumulating experimental evidence tells a more sobering story. Randomized evaluations across South Asia, Sub-Saharan Africa, and Latin America consistently reveal that enrollment gains have not translated into commensurate learning gains. The World Bank now describes this gap as a global learning crisis: hundreds of millions of children sit in classrooms daily without acquiring foundational literacy or numeracy.

This disconnect demands a rebalancing of development education priorities. Drawing on two decades of impact evaluations—from the Progresa/Oportunidades trials in Mexico to the Pratham teaching-at-the-right-level experiments in India—this analysis examines what rigorous evidence tells us about access interventions versus pedagogical reforms. The findings suggest that marginal dollars now yield substantially higher returns when directed toward learning quality rather than seat expansion, and that program design must confront the uncomfortable reality that schooling and education are not synonymous.

The experimental literature on access interventions is unusually rich, spanning multiple continents and intervention modalities. School construction programs, most famously evaluated by Duflo (2001) in Indonesia's INPRES expansion, demonstrated robust causal effects on educational attainment, with each additional school per 1,000 children yielding 0.12 to 0.19 additional years of schooling and measurable wage returns.

Fee elimination reforms, evaluated through natural experiments in Uganda, Kenya, and Malawi, produced dramatic enrollment surges—often 20 to 40 percent increases within a single year. Conditional cash transfer programs generated some of the most rigorously documented effects in development economics: the randomized rollout of Progresa in Mexico increased secondary enrollment by roughly 8 percentage points, with similar magnitudes replicated in Colombia, Nicaragua, and the Philippines.

Merit scholarships, girls' scholarships, and deworming interventions—the latter evaluated by Miguel and Kremer (2004)—have also produced statistically credible attendance effects, some at remarkably low cost per additional year of schooling induced.

The pattern across this evidence base is consistent: reducing the price of schooling reliably increases the quantity consumed. Where financial or physical barriers bind, marginal children respond to marginal reductions in cost. This is neoclassical demand theory operating exactly as predicted.

The critical limitation, however, lies in what these interventions measure. Enrollment, attendance, and years completed are inputs to human capital formation, not outputs. The experimental designs that convincingly identified access effects rarely measured learning outcomes with comparable rigor, creating a systematic blind spot in the field's understanding of program effectiveness.

Takeaway

Reducing the price of schooling reliably increases attendance—but attendance is an input, not an outcome. Confusing the two has quietly distorted a generation of development spending.

The most consequential empirical finding of the past fifteen years in development education is the systematic decoupling of schooling from learning. ASER surveys in India, Uwezo assessments in East Africa, and the World Bank's harmonized learning outcomes database converge on a disturbing pattern: majorities of children in grade five cannot perform grade-two level reading or arithmetic tasks.

In rural India, roughly half of grade-five students cannot read a grade-two text fluently. In Kenya and Tanzania, similar proportions of grade-three children cannot perform two-digit subtraction. These are not measurement artifacts—the assessments are simple, orally administered, and calibrated to basic curricular standards.

Pritchett's characterization of schooling ain't learning captures the analytical challenge. Impact evaluations of pure access interventions, when they include learning measurements, frequently find null or negligible effects on test scores despite substantial effects on enrollment. The marginal child brought into school by a conditional transfer often enters a system where the median child is already learning very little.

This finding has profound implications for cost-effectiveness analysis. Standard economic models value schooling because it produces cognitive skills that translate into productivity. If the production function from years of schooling to learning is broken, the returns to enrollment expansion collapse. J-PAL's cost-effectiveness rankings now consistently show that dollars spent on learning-focused interventions produce ten to hundred-fold greater learning gains per dollar than access-focused alternatives.

The learning crisis is not primarily a resource problem—many high-performing systems spend less per pupil than low-performing ones. It is a problem of pedagogical practice, curricular pacing, and accountability structures that experimental methods are uniquely positioned to diagnose.

Takeaway

The production function from schooling to learning is not automatic. When it breaks, expanding access without repairing pedagogy simply scales an ineffective process.

The experimental literature on learning-focused interventions has identified a small set of approaches with unusually robust effect sizes. Teaching at the Right Level (TaRL), developed by Pratham and evaluated in multiple randomized trials by Banerjee, Duflo, and collaborators, groups children by demonstrated competence rather than age or grade and targets instruction accordingly. Effect sizes routinely exceed 0.3 standard deviations—substantial by education research norms—at costs often below five dollars per child.

Structured pedagogy programs, which provide teachers with scripted lesson plans, aligned learning materials, and ongoing coaching, have demonstrated consistent positive effects across evaluations in Kenya, Liberia, and South Africa. The Tusome program in Kenya, deployed at national scale, produced learning gains of approximately 0.6 standard deviations in early grade reading.

Teacher incentive experiments have yielded more heterogeneous results. Muralidharan and Sundararaman's work in Andhra Pradesh found positive effects from performance-linked pay, while other trials have found null or negative effects when incentives distort teaching toward narrow test preparation. The lesson is that incentive design interacts critically with measurement systems and teacher capacity.

Technology-assisted learning, particularly adaptive software that adjusts to individual student levels, has shown promising results—Muralidharan, Singh, and Ganimian's evaluation of Mindspark in Delhi found effects of roughly 0.37 standard deviations in math over four and a half months.

The unifying principle across successful pedagogical interventions is alignment between instruction and the actual learning level of students. Curricula in most developing country systems are paced for the top quartile of learners, leaving the bottom half progressively further behind each year. Interventions that break this pattern—whether through remedial grouping, adaptive technology, or structured teacher support—consistently outperform interventions that simply add more of the existing system.

Takeaway

Effective pedagogy meets learners where they are, not where the curriculum assumes they should be. The gap between these two points explains much of the global learning crisis.

The experimental evidence accumulated over two decades supports a clear rebalancing of development education priorities. Access interventions retain value where enrollment gaps remain large, particularly for marginalized populations and secondary schooling. But the marginal returns to further access expansion in systems with high enrollment and low learning are now demonstrably lower than the returns to pedagogical reform.

This rebalancing requires uncomfortable choices for donors and ministries whose institutional identities are tied to enrollment metrics. It requires learning measurement systems that generate actionable data at the classroom level, curricular reforms that abandon aspirational pacing, and teacher support structures that make effective instruction feasible rather than heroic.

The evidence-based development agenda has matured to the point where we know a great deal about what works. The remaining challenge is institutional: designing delivery systems that can implement pedagogical interventions at scale without eroding the mechanisms that made them effective in trial conditions.