Effects of COVID-19 education policies on students’ learning proficiency in Latin America—Insights from PISA 2022

Latin America’s education systems – with a high prevalence of students not achieving the basic competencies, entrenched inequalities and weak connectivity, and longer schools closures – were highly affected by the COVID-19 pandemic, experiencing sizeable learning losses. The role of education policies to boost students’ engagement for remote learning were critical during schools shutdowns. Yet, there is a lack of evidence on whether pandemic educational policies were effective. Using explainable machine learning methods and a sample of ten countries from PISA 2022, this paper fills this gap by analysing the association of three education policies for remote learning (students’ resources, teachers’ resources and teachers’ communication) implemented during the pandemic with the probability of students reaching competency (level 2) in math and reading. I find that policies’ effects on learning are non-linear with specific thresholds where policy become more efficient to keep students’ chances of level 2 achievement, with these thresholds linked to schools’ length of closure and the intensity of policies. I also find that policy intersectionality is important with family wealth and educational inputs being pathways to counteract associations under longer schools’ closures, and policies not being capable of narrowing learning loss by gender (against girls). Results by educational sub-systems for the policy on students’ resources indicate that by school type there is similarity on associations for reading, but for math this policy is more productive in private than public schools, as for students from urban schools under longer period of closure than students from rural schools. • Assess the link of pandemic educational policies with learning for 10 LAC countries. • Remote learning policies analysed: students’ and teachers’ resources and communication. • Find that policies’ effects on learning are non-linear with specific thresholds. • Results vary by students’ wealth, gender and by school type and location. • Analysis is based on explainable machine learning approach: the SHAP method.