Local processors and global integrators: Identifying L2 reader profiles using mixture item response theory models.
L’essentiel
This study aims to explore multiple profiles of second/foreign language (L2) readers by applying several mixture item response theory (MixIRT) models to the reading comprehension section of a high-stakes multiple-choice language test. The study characterizes the classes based on examinees' gender, lexico-grammatical knowledge, and overall language proficiency, measured by a Cloze test. Item responses of 2439 examinees to the reading comprehension section of the test were analyzed using a range of MixIRT models, including the mixture Rasch model, two parametric logistic MixIRT (2PL MixIRT), 3PL MixIRT, and 4PL MixIRT, with one to six latent classes. The 2PL IRT model with two classes showed the best fit to the data. The two classes were: (1) Local Processors and (2) Global Integrators. Class 1 comprises lower- to moderate-level proficiency examinees who possess restricted overall language proficiency and lexico-grammatical knowledge and rely on bottom-up, sentence-level processing, and superficial strategies such as memorizing isolated lexical and grammatical forms. However, Class 2 involves higher-proficiency examinees who have higher general language ability and lexico-grammatical knowledge and coordinate top-down and bottom-up processes, integrate higher- and lower-level (sub)skills, and adopt predictive and inferential strategies for coherently understanding a text.
Synthèse détaillée
Résumé original
This study aims to explore multiple profiles of second/foreign language (L2) readers by applying several mixture item response theory (MixIRT) models to the reading comprehension section of a high-stakes multiple-choice language test. The study characterizes the classes based on examinees' gender, lexico-grammatical knowledge, and overall language proficiency, measured by a Cloze test. Item responses of 2439 examinees to the reading comprehension section of the test were analyzed using a range of MixIRT models, including the mixture Rasch model, two parametric logistic MixIRT (2PL MixIRT), 3PL MixIRT, and 4PL MixIRT, with one to six latent classes. The 2PL IRT model with two classes showed the best fit to the data. The two classes were: (1) Local Processors and (2) Global Integrators. Class 1 comprises lower- to moderate-level proficiency examinees who possess restricted overall language proficiency and lexico-grammatical knowledge and rely on bottom-up, sentence-level processing, and superficial strategies such as memorizing isolated lexical and grammatical forms. However, Class 2 involves higher-proficiency examinees who have higher general language ability and lexico-grammatical knowledge and coordinate top-down and bottom-up processes, integrate higher- and lower-level (sub)skills, and adopt predictive and inferential strategies for coherently understanding a text.