From Averages to Individuals: The Next Frontier of Evidence-Based Education

Over the past decade, the landscape of education policy has undergone a quiet revolution. We have moved away from “gut feelings” about what works best for children toward a rigorous, evidence-informed approach. The rise of the Randomized Controlled Trial (RCT)—often considered the “gold standard” for causal evidence—has been the engine of this change. Numerous foundations and organizations are now building the evidence needed to improve learning for all children. For example, in the UK, the Education Endowment Foundation has commissioned over 150 large-scale RCTs, reaching approximately 1.8 million children. In the US, the Institute of Education Sciences has funded hundreds of trials through its “What Works Clearinghouse,” transforming how districts choose reading programs. Similarly, in the global development world, organizations like the Abdul Latif Jameel Poverty Action Lab and the World Bank have used RCTs across Africa and Asia to identify effective interventions, such as “Teaching at the Right Level” (TaRL).

This progress is monumental and a testament to the funders and researchers who have spent years building an evidence base of “what works.” But as our evidence base grows, we are hitting a “resolution problem”: by focusing on effectiveness for the average child, we risk overlooking individual variability.

The Problem with “On Average”

RCTs tell us if a program works on average; by design, these averages aggregate across learners to look for patterns in the group. While valuable, an average treatment effect can act as a mask. For instance, a trial might show that a math app helped a group of 1,000 students, but that single number hides the fact that 300 students thrived, 400 stayed the same, and 300 actually fell behind.

Even when RCTs attempt subgroup analysis—looking at boys vs. girls or low-income vs. high-income students—these snapshots are often underpowered and too coarse to capture the true complexity of learning. Moreover, because large-scale RCTs are costly to run, they are typically limited to a small number of outcome measures; the cost of administering in-depth assessments to every child is simply too high. This results in us seeing a final score while often missing the “how” and the “why.” To move toward a system that optimizes for each child, we need to move toward tools and metrics that can speak to the individual.

Where LEVANTE can help

LEVANTE isn’t an intervention study; it’s a global research network dedicated to capturing a wide range of developmental markers across diverse contexts. By building the infrastructure to understand individual variability, LEVANTE is paving the way for a more nuanced era of education policy through three primary avenues.

First, LEVANTE is establishing the basic science of variation. The project allows us to understand the multidimensional profiles of skills that children carry. Instead of seeing a “struggling reader,” LEVANTE helps us see a child with high executive function but low phonological awareness living in a specific linguistic environment. By understanding how these individual traits and environments interact over time, we provide policymakers with a blueprint of where variability comes from before an intervention even begins.

Second, LEVANTE is developing the scalable measurement tools needed to modernize how we evaluate interventions. As noted above, a limitation of traditional RCTs is that measurement is often a “one-shot” affair: it can be expensive, limited to just one to three measures, and confined to a simple baseline and endline. As a result, trials rarely capture how an intervention shapes a child’s development months or years later. LEVANTE addresses this by validating tools that are efficient, cross-cultural, and rigorous enough to be integrated into future large-scale trials. By combining the causal strength of RCTs with high-resolution, repeated measurement, we can move toward tracking individual learning trajectories over time.

Finally, although not an endeavor we would undertake anytime soon, LEVANTE has the potential to serve as a platform for testing interventions—whether through traditional RCTs or alternative evaluation methods. Because the project provides a rich, longitudinal baseline of individual variability, it could one day allow researchers to tailor a suite of targeted, online interventions based on a child’s unique performance across our nine core tasks. For example, if the data shows a child is excelling in vocabulary but struggling with numeracy, the platform could potentially deliver a specific module to assist with math learning. This approach would move us away from more blanket learning models toward a system tailored to the specific needs of the child. 

In short, the evidence revolution in education has brought us a long way—but it has also revealed its limits. If we want policy and practice to work not just on average but for each child, we need a sharper lens on how learning varies across individuals and contexts, and better tools to measure change over time. By building the basic science of variability and the scalable measurement infrastructure to capture developmental trajectories, LEVANTE helps bridge the gap between robust causal evaluation and real-world personalisation—so the next generation of “what works” can also answer for whom, when, and why.