Applications of learning dynamics in developing countries: Modelling policies for improving learning

• In Event: Instructional (In)coherence: How greater alignment between curriculum, instruction and children’s learning levels can improve student learning

Proposal

Learning outcomes vary dramatically, both across and within countries, regions and classrooms (Pritchett and Sandefur, 2017; Muralidharan and Singh, forthc.). These differences in learning outcomes are the result of differences in children’s learning trajectory or learning process. While it is important to focus on learning outcomes, the process that leads to these observed outcomes is complex and often neglected. Predicting how changes in the learning process will impact learning outcomes requires an understanding of the parameters at work and the ways they interact. Contributing to a structured understanding of the learning process, we present the Pedagogical Production Function (PPF), a parametrized, structural model to capture the defining features of the learning process and its interactions. The PPF also has a variety of real-world policy applications for improving learning.

Based on a synthesis of the existing empirical literature on learning profiles, the PPF consists of a clear set of parameters that formally characterize the learning process and that can be calibrated to reproduce the distribution of a variety of observed learning outcomes. In a first round of simulations, it was calibrated to replicate the 2018 PISA for Development (PISA-D) results of seven low- and middle-income countries. However, the PPF parameters are flexible and can be calibrated to other datasets.

By simulating different policy options, the PPF allows for comparisons between potential long-term effects of these options on learning outcomes. For instance, the PISA-D calibrated model shows that expanding schooling to universal grade 10 completion would produce very little additional learning. As such, continuing with the business as usual focus on access—without simultaneous action to improve learning—would not raise the percentage of children reaching the learning goal of SDG 4.1. Adjusting other parameters in the model, however, has large, positive effects. Slowing the pace of curriculum, so that more children can keep up, increases average learning in grade 10 by the learning equivalent of 1.6 years of schooling. Improving classroom instruction and extending instruction towards the skill levels of lower performers yield learning gains of 1.8 and 1.0 years respectively.

The PPF can also integrate learning shocks. Using the same calibration, the PPF can be applied to simulate the potential long-term learning losses caused by the COVID-19 related school closures and the effects of different mitigation strategies on learning outcomes (Kaffenberger, 2020).

Projections like these can help policymakers estimate the magnitude of change for different scenarios. At the same time, not all policy levers may be socially desirable, politically implementable or financially feasible. The PPF allows to keep certain parameters constant while optimizing others, building different reform options based on a specific political environment. With all its application options and flexible design, the PPF has the potential to become a more reliable guide to policy for improved learning.

Authors

• Carmen Belafi, RISE Programme

• Michelle Kaffenberger, RISE Programme

• Lant Pritchett, Blavatnik School of Government, University of Oxford