Linguistic Complexity of the Assessments

• In Event: Students’ Diversity and Test Fairness

Abstract

Assessments that are developed and field tested for mainstream native English speakers may not provide valid and fair assessments for English language learner (ELL) students. A major threat to the validity of assessment for ELL students is the impact of language factors on their assessment outcomes. ELL students often perform significantly lower that their native English speaking peers particularly on content areas with high level of language demands. Therefore, the assessment content and linguistic structure of the assessments are said to be confounded (see, for example, Abedi, 2010; Abedi, 2009). An ELL student may have the knowledge of the content covered in the test but may not be at the level of English proficiency to understand test questions or write responses to the questions.
In the studies of the impact of language factors on the assessment of ELL students a distinction is made between language that is relevant to the focal construct (relevant to the construct being measured) and language that is judged to be irrelevant to the construct (construct irrelevant). The decision on relevance of the linguistic structure of the assessment is made by a group of experts including content, linguistic and assessment experts. The linguistic complexity due to irrelevant sources is then revised and a more linguistically accessible assessment version is presented to ELL students (see Abedi & Lord, 2001; Abedi, 2006; Maihoff, 2002; Kiplinger, et al, 2000; Young, 2008).
In this symposium, we will present studies that used randomized field trail design to investigate the impact of unnecessary linguistic complexities on the content-based assessments (such as mathematics and science). For examining the impact of linguistic complexity on the assessment outcomes we obtained currently released test items from the national (e.g. NAEP) and state assessments in mathematics and science content areas and indentified features of linguistic complexities based on the rubric developed and validated in the literature (Abedi, 2006). We then linguistically revised items that have higher complexity ranking (4 and 5 ranking in a 5-point Likert scale) and created two version of the test, an original version with some complex linguistic structure and a revised version in which unnecessary linguistic complexities were removed or recast.
Students were then randomly assigned to take either the linguistically revised version of the test (the treatment group) or the original version (the control group). By random assignment of the original and revised version of the mathematics and science items to students (grade 8), we controlled for the possible initial differences between the two groups so that any significant differences could be attributable to language differences between the two forms.
Results of several studies that will be presented in this symposium consistently suggested that ELL students perform significantly better under the revised version of the test in which unnecessary linguistic complexity in the test items was adjusted.

Author

• Jamal Abedi, University of California - Davis