Estimating the Learning Loss of Students from School Closures due to the COVID-19 Pandemic: A Bayesian Piecewise Regression Analysis

Abstract

Estimating learning loss is vital information for education, as it can guide policymaking to prescribe a policy for solving or preventing the issue. This study aimed to achieve two objectives: (1) to design the estimation of learning loss among students in schools, and (2) to analyze the learning loss among students in schools using Bayesian Hierarchical Piecewise Regression (BHPR) modeling. The analysis utilized O-NET mathematics scores of sixth-grade students from the year 2017 to 2022 from schools under the supervision of the Basic Education Commission, totaling 1,707 schools. The research found that estimating learning loss among students in schools using BHPR analysis provided estimates consistent with the trend of O-NET scores of students in schools. This is because parameters showed the relationship between O-NET scores and academic years, as well as the cutoff points before and after the outbreak. Among all 1,707 schools, 346 schools (25.42%) had the highest density range of 89% outside the ROPE range, indicating significant learning loss. Large special-sized schools located in urban areas were found to have the highest learning loss scores, with an average score of -6.26 (SD = 3.41). Conversely, other schools with significant learning loss scores had average scores ranging from -1.69 to -3.49. However, when considering individual schools, the school with the highest learning loss was a small-sized school located outside the city, with a learning loss score of -12.73.