BLAVATNIK SCHOOL OF GOVERNMENT HOW TO IMPROVE EDUCATION

When we compare studies using standard deviations as our metric, we impose the assumption that the difference in learning levels between the median and 66th-percentile student in a fourth-grade math class in Kenya is equivalent to the difference in learning levels between the median and 66th-percentile student in a twelfth-grade history class in Peru. [...] If $100 buys an additional 6 months of schooling for a child, is that a good buy if the quality of schooling is bad? Is $100 for an increase in test scores of 0.05 standard deviation a good investment? The answers depend on the 4 underlying quality of the additional schooling in the first case and on the underlying heterogeneity in learning outcomes in the second. [...] If we assume that students will stay in school once the school is built and that the quality of schooling remains constant, we can then adjust this estimate by the remaining school life expectancy (i.e., the number of grades in a given school system minus the grade at which the students entered the school), because we expect students to continue to benefit even after the evaluation period. [...] We derive equivalent years of schooling, e, by expressing learning gained relative to learning in the status quo: βtest e = i δtesti where β is the learning gain produced by the intervention per year in country i; test denotes the test used to measure learning; and δ is the status-quo learning rate per year in country i.7 The student sample for the status quo learning trajectory, δ, in country i c. [...] In the main analysis reported in the paper, we use national-level learning trajectories; in this section, we explore the alternative of using the study’s control group to measure status-quo learning.