An investigation of item fit and functioning of an algebra assessment for English language learners.

摘要

Important decisions are often made based on assessment results, but there are questions about whether they really assess what is intended for all students, including English Language Learning (ELL) students. This study asks the following questions: (1) Do the data from a multiple choice assessment designed to measure the algebra ability of a group of high school students fit the requirements of the Rasch model for both ELL and English-proficient students? (2) How do the item difficulties compare for ELL and English-proficient students? (3) If certain data fit the model poorly and/or have divergent difficulty measures for the ELL students and their English-proficient classmates, can these idiosyncrasies be attributed to the linguistic complexity of the items?;The Rasch model was selected because deviations from the expectations of the model may indicate problematic features that can be controlled for or explained, especially when they are related to a demographic subgroup such as ELL students. Furthermore, it supports the relatively small sample of ELL students included in the study. Through the application of the Rasch model, along with a systematic method of selecting an item set varying in linguistic complexity and mathematical content difficulty, this study provides a look at the role linguistic complexity plays in determining item difficulty for ELL students on a mathematics assessment.;Empirical evidence revealed that proportionately more ELL students misfit than English-proficient students. Surprisingly, poor fit resulted from many ELL students answering certain difficult items correctly, and all but one of these items required students to translate a verbal expression into a mathematical one. The five differentially difficult items for the ELL students contained a key word or phrase important to understanding the item, and contained material that students new to the United States may not have had the opportunity to learn.;The findings challenge current notions about linguistic complexity confounding content assessment, although further studies should be conducted before definitively countering others' work. This study offers new ideas to consider, and a methodology for conducting the future studies to further explore these ideas.