Abstract
<p>Spanish-English bilinguals are the largest bilingual group in U.S. schools, yet the frameworks used to predict their reading from early screening data were developed on largely monolingual samples, leaving open the question of which language(s) to assess. This study compared three approaches to predicting end-of-year (EOY) reading from beginning-of-year (BOY) early-literacy and language skills in 553 Spanish-English bilingual kindergarten and first-grade students: English-only, Spanish-only, and a derived best-language score (the higher of a child's two single-language scores per task). Linear mixed-effects models, controlling for school clustering, compared these approaches separately by grade, outcome language (English/Spanish), and instructional context (bilingual vs. English-only). Best-language scoring did not uniformly improve prediction; benefits diminished with grade. In kindergarten, it matched or modestly exceeded the best single-language approach (marginal R² = .35 for English outcomes, .35 for Spanish outcomes) and explained the most variance for Spanish outcomes. By first grade, once target-language word reading emerged as a dominant proximal predictor, best-language scores offered no advantage over English-only assessment for English outcomes (R² = .57 vs. .43). English-only assessment predicted Spanish reading poorly, especially in kindergarten, while Spanish and best-language approaches predicted both outcome languages reasonably well throughout. Findings align with a distributed-knowledge account: decoding-related skills transfer across languages and benefit from best-language scoring, while language-comprehension skills are language-specific and gain little from it. Practically, assessment language should be matched to the skill, grade, and outcome of interest rather than defaulting to one language for all children.</p>