Off-campus UMass Amherst users: To download dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users, please click the view more button below to purchase a copy of this dissertation from Proquest.

(Some titles may also be available free of charge in our Open Access Dissertation Collection, so please check there first.)

Using a mixture IRT model to understand English learner performance on large-scale assessments

Christine A Shea, University of Massachusetts Amherst

Abstract

The purpose of this study was to determine whether an eighth grade state-level math assessment contained items that function differentially (DIF) for English Learner students (EL) as compared to English Only students (EO) and if so, what factors might have caused DIF. To determine this, Differential Item Functioning (DIF) analysis was employed. Subsequently, a Mixture Item Response Theory Model (MIRTM) was fit to determine why items function differentially for EL examinees. Several additional methods were employed to examine what item level factors may have caused ELs difficulty. An item review by a linguist was conducted to determine what item characteristics may have caused ELs difficulty; multiple linear regression was performed to test whether identified difficult characteristics predict an item's chi-squared values; and distractor analysis was conducted to determine whether there were certain answer choices that were more attractive to ELs. Logistic regression was performed for each item to test whether the student background variables of poverty and first language or being an EL predicted item correctness. The DIF results using Lord's Chi-Squared test identified 4 items as having meaningful DIF >0.2 using the range-null hypothesis. Of those items, there were 2 items favoring the EO population that were identified as assessing the Data Analysis, Statistics and Probability strand of the state Math Standards. As well, there were 2 DIF items that favored the EL population that were identified as assessing the Number Sense and Operations strand of the state Math Standards. The length of the item as judged in the item review supported several items that were identified as DIF. The Mixture IRT Model was run using 3 conditions. It was found that with all three conditions, the overall latent class groupings did not match those of the manifest groups of EO and EL. To probe further into the results of the latent class groupings, the student background variables of poverty, language proficiency status or first language spoken were compared to the latent class groupings. In looking at these results, it was not evident that these student background variables better explain the latent class groupings.^

Subject Area

Educational tests & measurements

Recommended Citation

Shea, Christine A, "Using a mixture IRT model to understand English learner performance on large-scale assessments" (2013). Doctoral Dissertations Available from Proquest. AAI3603151.
http://scholarworks.umass.edu/dissertations/AAI3603151

Share

COinS