Guatemala's Scalable Early-Warning System for School Dropout Prevention

Guatemala has one of the highest school dropout rates in Central America: nearly 40% of sixth-graders fail to complete ninth grade. Working with the Ministerio de Educación (Mineduc), World Bank researchers Francisco Haimovich, Emmanuel Vázquez and Melissa Adelman built a statistical prediction model that uses only data schools already collect — prior grades, attendance, age-for-grade, and school characteristics — to flag, before the school year began, which sixth-grade finishers were most likely to drop out during the transition to lower-secondary school. Because the model reused an existing government information system rather than requiring new data collection, its marginal cost was negligible.

To identify at-risk sixth-grade finishers early enough for schools to intervene with targeted outreach and support before they drop out during the transition to lower-secondary school, using only existing administrative data so the tool could scale nationally at near-zero marginal cost.

Flagged students were referred to school staff for targeted outreach and support. The approach was tested in a randomized controlled trial across 4,000 primary schools (about 17% of Guatemala's primary schools) in 2018, registered as AEA RCT ID AEARCTR-0004091.

Schools assigned to the programme saw a 4% relative reduction in dropout (1.3 percentage points) during the primary-to-secondary transition; among schools that implemented the intervention as designed ('compliers'), the reduction reached 9% (3 percentage points). Estimated programme cost was under US$3 per student. The World Bank and researchers describe the effect size as 'relatively modest', while noting the near-zero marginal cost and reliance on existing administrative data make this a rare case of a predictive education tool tested at national administrative scale with a genuine control group.

Because the model depends entirely on the completeness of administrative records, its accuracy is bounded by the underlying school data systems; the World Bank has separately worked with Guatemala to strengthen these statistical capacities. A parallel prediction model was co-developed for Honduras using the same method.