World Development Report 2018: Learning to Realize Education’s Promise

As evidence of the learning crisis has grown, so has translates into clearer insights into how to improve understanding of what produces learning. Cognitive learning at the level of the student, the classroom, and the school. Beyond the increase in their numneuroscience has evolved dramatically, with brain ber, these impact evaluations have also grown more imaging revealing new insights into how children sophisticated over time, making them more useful for learn. Over the last two decades, neuroscience has policy making. They are now more likely to compare been instrumental to understanding early child brain multiple interventions, more likely to study a wide development and the crucial nature of the early years.

range of interventions overall, and more likely to

Schools in many parts of the world are innovating in study interventions on a large scale. The evaluations approaches to pedagogy, professional development, show that many of these interventions have sizable and the use of new technologies. Governments and impacts. Several pedagogical interventions, for examnonprofits are trying out innovative programs to ple, deliver learning gains greater than what students upgrade teachers’ skills on the job.

would learn in a year of business-as-usual schooling.

At the same time, evidence on which programs most effectively boost learning is mushrooming. One example of that growth: the number of impact evaluations of interventions intended to improve learning outcomes in developing countries rose from

Not all evidence is created equal, but many different

19 in 2000 to 299 by 2016 (figure S4.1). This evidence kinds of evidence can be credible. Scientific evidence demonstrates the pathways of brain development and functioning. Social science evidence can effectively answer the question of what would have happened in the absence of a reform or intervention (often called a counterfactual). Randomized controlled trials or analyses of “natural experiments” are useful tools for determining such a counterfactual. Implementation science and case studies can provide a detailed picture of how an intervention or a phenomenon works. The best evidence of what improves learning draws from a range of methods.

Even when an intervention in one education system has a positive impact, it may not work everywhere. Effects may differ when translating from one location to another or from a pilot study to a large-scale program. What works in Peru may not work in Burundi because the education systems and societies are different. A common intervention that has been tested in a range of settings is to reduce class size. But increasing class size by 10 students reduced test scores by four theory—models of human behavior—to explain why times as much in Israel as it did in Kenya. A pilot intersome proposed solutions work and others do not, as vention may allow for more controlled conditions than well as why the same solution may work in one locale an at-scale intervention. In Kenya, an intervention to or time but not in another.

hire contract teachers was effective on a small scale, but when it was implemented at scale through government systems, salaries were delayed and ultimately the contract teachers were converted to civil servants. The scaled-up program no longer resembled the successful pilot, and the learning gains failed to materialize.

To make sense of the evidence, policy makers should consider the likely principles behind effective programs rather than fixating on results (or “point

Many actors contribute to the learning process, and estimates”) from individual studies. For example, prothey all face their own incentives. The direct inputs grams that provide financial incentives for teachers to the learning process include the choices made by have had mixed effects. Rather than taking a simple learners themselves, as well as by their parents, teachaverage of the effects, a nuanced assessment would ers, and other school leaders, interacting with the reveal that these programs tend to work better when available infrastructure and materials. Less immediimproving quality is relatively simple and within a ate but still important, bureaucrats, politicians, and teacher’s control—for example, when they increase nonstate players make decisions that influence eduteacher attendance or teaching time while at school.

cation quality. Understanding these relationships is

Viewing evidence through models of human crucial to interpreting evidence.

behavior is one way to focus on principles. This

Each actor in the learning process reacts to the means examining patterns of results and using modothers, so changing one element of the process does els to infer why results vary across settings. The first not guarantee more learning. Many of the inputs to step would be a nuanced synthesis, bringing together the learning process are choices made by the actors— the results of a range of studies and examining choices made in reaction to the actual and anticipated empirical patterns. The second step would be using choices of other actors (figure S4.2). Teachers react to changes in school leadership, school directors react to community demands, and parents react to changes in government policy. In India and Zambia, grants to schools led parents to reduce their own investments in their children’s schooling. In a household with few resources, if the government begins providing textbooks, a parent may well reallocate education resources to other needs, such as health.

How can we make sense of all of these complex, dynamic relationships? Models of human behavior illuminate the motives for choices and actions, and they can help guide solutions. Simple optimizing behavior models—in which actors maximize their well-being subject to limited budgets and other constraints—explain why parents reduce their contributions when schools increase theirs. Principal-agent models that incorporate multiple actors with different objectives explain why teachers may fail to teach when not sufficiently motivated or monitored. Behavioral models also play a role: student learning and educational aspirations can be affected by the salience of stereotypes. Economic phenomena such as information, market, and coordination failures play a role in these models. The models can also illuminate why a gap is often observed between evidence on how to improve learning and actual practice.

Gaps between evidence and actual practice provide entry points for efforts to improve education. These gaps come to light when evidence shows that certain approaches or interventions can improve outcomes, but the approaches used in practice are different. For example, the accumulated research evidence demonstrates high returns to early investments in children, yet families and governments in low-income environments do not prioritize these investments. Evidence shows that certain types of teacher professional development deliver much higher learning gains than others, but outdated training methods persist. Because the gap between evidence and practice requires good information on what the evidence says, as well as what current practice is, it is likely that many opportunities for improvement have yet to be discovered.

Intuition and common sense are not enough. One fundamental lesson from the growing evidence base is that intuition is not always a trustworthy guide. It may miss the complexity of motivations and reactions in the real world, as can happen when teacher financial incentives induce cheating rather than more effort. Intuition may fail to capture the net effect of conflicting forces, such as when separating students by ability allows teachers to target teaching more specifically to students’ level—which should increase their learning—but also distances them from their high-performing peers—which may decrease their learning.

Knowledge about improving learning must take both the costs and the benefits of learning interventions into account. A computer-assisted learning intervention in India increased learning more than employing contract teachers in Kenya, but hiring contract teachers was so much cheaper that it delivered a higher return on investment. The evidence base on costs is much thinner than that on benefits, with a tiny fraction of studies examining both. But some programs have been evaluated on both effectiveness and cost-effectiveness. This evidence on costs—adapted to local contexts—should qualify policy recommendations.

The gaps between evidence and practice signal promising places to start, rather than the end of learning how to improve learning. Interventions cannot simply be exported from one country to another. Indeed, at times the effectiveness of an apparently similar intervention can vary even within a country, depending on how the program is implemented. The cost of implementation will also vary dramatically across contexts. But this does not mean that evidence from other contexts is without value. On the contrary, successes in other environments—coupled with a careful analysis of why the programs work— provide a starting point. Policy makers can draw on this evidence and experiment in their own policy environment.