A couple of weeks ago, we set the stage for developing evidence-based, data-driven answers to questions about learning loss resulting from COVID-19 related school closures. We have now analyzed 27 existing datasets from low and middle income countries (LMICs) to estimate year-on-year growth in student reading achievement under normal conditions, as a starting point for developing models to estimate learning losses (relative to expected growth). Since learning loss will not be equal for all students, we focus on examining full distributions of scores in order to produce more precise estimates than could be obtained by relying only on average change. These analyses have yielded promising evidence of common trends across countries, grades, and languages—leading us toward a new model of estimating loss.
How the model works
We began by examining distributions of oral reading fluency scores from students in consecutive grades (e.g. grade 1 and grade 2; grade 2 and grade 3). The idea is that grade 3 scores are our best prediction of where grade 2 students will be after one year of schooling. Therefore, we can use this approach to model a full year’s learning gain or conversely an expected loss caused by school closures and disruptions. An example from the Philippines is displayed in Figure 1.