Decision Rules for Identification of Risk for Reading Difficulty

Abstract

Early identification of students at risk for reading difficulty (RD) is central to multi-tiered systems of support, yet existing screening tools often show only poor to moderate accuracy, particularly for longer-term outcomes. This study evaluated a novel screening approach designed to address these limitations through two complementary innovations. First, school-level contextual indicators (e.g., RD base rates and English learner composition) were incorporated to capture environmental influences without increasing testing burden. Second, a Classification and Regression Trees (CART) framework was used to produce an interpretable model with clear decision rules for practitioners. Results indicated that contextual variables, particularly school-level RD base rates and socioeconomic composition, contributed meaningfully to classification, supporting theories emphasizing the role of educational context in reading development. However, the model did not meet established benchmarks for diagnostic accuracy and did not perform consistently across demographic subgroups. These findings highlight ongoing challenges in developing accurate screening tools. Future research should identify additional contextual indicators, address measurement error in decision cutpoints, and better balance model fit with generalizability. Although not yet ready for implementation, this study provides initial support for a CART-based, context-informed approach to screening development.