Institute for Environmental Research and Education

This study examined the genre‑based progression of writing competence from 5th to 7th grades of Ecuador’s Basic Education across contrasting settings: Guayas (coastal, more urban) and Chimborazo/Cotopaxi (Andean, more rural/indigenous). We employed an explanatory sequential mixed‑methods design (QUAN→qual) combining: (i) primary sampling of authentic student texts in four school genres (narrative, news report, description, procedural/instruction), scored with analytic rubrics (α/κ ≥ 0.70; metric invariance across provinces); (ii) extraction of NLP metrics (lexical density, T‑unit length, referential cohesion, connectives); and (iii) linkage with open data (SEST‑INEVAL grade‑7 scores; AMIE; H3/OSM/WorldPop/VIIRS). Analyses included multilevel models (student–section–school), Genre Progression Index (GPI) estimates, and school‑level correlations with SEST. Findings indicate a substantive 5th→7th progression that varies by genre, with steeper gradients in news and procedural writing. Guayas shows a relative advantage in news (higher GPI and grade slope), tied to improvements in inverted‑pyramid structure, lead quality, and source use; the Andean sample displays qualitative strengths in register and audience alignment. NLP metrics partially mediate the grade effect (indirect β ≈ 0.06), adding sensitivity to textual complexity. At the school level, news writing (grade‑7) is the strongest predictor of SEST Language & Literature (β ≈ 0.31), followed by procedural writing (β ≈ 0.24). Robustness held under design weights, multiple imputation, and alternative specifications. We conclude that genre‑based instruction—making macrostructure and cohesion explicit through mentor texts and formative feedback—accelerates writing development. We recommend hybrid sequences (deconstruction→joint writing→independent writing), emphasis on news/procedural genres, and low‑cost kits (templates and open rubrics) prioritizing rural and EIB contexts. The open‑data package (anonymized microdata, scripts, JSON‑LD metadata) enables replication, auditing, and responsible scaling