SGER: Teacher-adapted versus linguistically simplified items in the testing of English language learners

American Institutes For Research In The Behavioral Sciences, Arlington VA

Investigators

Abstract

This project addresses the validity of academic achievement measures for English language learners (ELLs) and the need for effective approaches to testing these students. In this study, we will examine the dependability of measures of mathematics achievement produced by two forms of accommodation for ELL students: linguistic simplification and teacher adaptation. In the former, a team of professionals modifies the wording of the items with the intent to reduce their linguistic demands; then a single set of modified items is given to all students in the target population of ELLs. In the latter, teachers modify the wording based on their knowledge of the language used in their communities; then students from each community are given local versions of the adapted items. While effective, linguistic simplification has produced only modest results in reducing the impact of language on ELL performance on tests. The effectiveness of teacher adaptation has just begun to be investigated. Evidence thus far shows that this promising form of accommodation is sensitive to subtle but important variations due to language. In this project, we will assemble a test composed of 30 mathematics items. From this original version of the test, we will create two versions, linguistically simplified and teacher-adapted. ELL students who are native speakers of Spanish will take the same set of items in two of these three versions (original and linguistically simplified, original and teacher adapted, or linguistically simplified and teacher adapted). We will assess the effectiveness of the two forms of accommodation by examining the differences between the mean scores obtained by the students on the linguistically simplified and teacher-adapted test versions and between each of these versions and the original version of the test. Also, based on generalizability (G) theory "a theory of measurement error" we will assess the dependability of the scores obtained with each test version and the number of items that are needed with each test version in order to be able to make valid generalizations about the students' knowledge of the domain.