SIGES Alerta Temprana — El Salvador's Self-Audited School Dropout Early-Warning System

El Salvador's Ministry of Education, Science and Technology (MINEDUCYT) built an early-warning module into SIGES, the country's national student information system, to flag students at risk of dropping out. The module cross-references each student's record — over-age-for-grade status, absences, academic performance, teenage pregnancy, early parenthood, early union or marriage, and exposure to violence — to classify dropout risk as medium, high or very high and suggest school-level interventions. UNICEF financed a 2020 diagnostic and methodology-validation phase, piloted with roughly 300 teachers and administrators through the NGO FEDISAL.

The system aims to identify students at risk of dropping out early enough for schools to intervene, using administrative data already collected in SIGES rather than a separate data-collection exercise. Its risk classification is rules-based, drawing on indicators such as pregnancy, early union, over-age status and absenteeism, rather than a trained predictive model.

In 2021–2022, MINEDUCYT's monitoring department and a UNICEF-funded consultant (FEDISAL) carried out a national data-quality audit, manually verifying SIGES's automated alerts against door-to-door and paper-form data collection for the 2021 school year. The verification exercise covered 1,224,249 students across 5,971 public and private education centers, with 96.38% of centers completing the process.

The audit found the automated system dramatically undercounts true at-risk cases: SIGES recorded 172 pregnancy alerts versus 1,348 found through field verification (about 8 times more), 831 underage-parent alerts versus 8,652 in the field (about 10 times more), and 1,995 early-union alerts versus 7,547 in the field (about 4 times more). No outcome evaluation of whether flagged students' dropout rates actually improved has been published.

This gap means the automated early-warning system, as deployed, likely misses the large majority of the vulnerable students it is designed to identify — a serious, self-documented limitation rather than an outside criticism. The audit's transparency is nonetheless notable: few government AI-adjacent administrative tools have had their failure modes so openly quantified and published by the implementing ministry and its funder.