Abstract
This study compares two approaches to grammatical knowledge in language assessment: the structural view that regards grammatical knowledge as vocabulary and syntax (Bachman 1990), and the communicative view that perceives it as the binary combination of grammatical form and meaning (Purpura 2004). 1,491 second-year nursing students from eight medical colleges in China took a fifteen-item English grammar test (GT) that used retired items from the Language and Use section of the Public English Test System, Level Two. Data analysis comprised a series of dimensionality assessments (DAs) based on bifactor- multidimensional item response theory (bifactor-MIRT). This involved assessing the model fit achieved by structuring the GT tasks using a single grammatical factor, the structural approach and the communicative approach, and then comparing the relative performance of the three approaches. The results indicated that 1) despite its operational attraction, a unidimensional structure was insufficient to structure the GT tasks; 2) both the structural and the communicative approaches could sufficiently explain the underlying structure of the GT tasks; but 3) the communicative approach seemed to outperform the structural approach in uncovering the factual structure of the GT tasks. The study shows how bifactor-MIRT can be used to compare grammatical knowledge theories.
About the author
Having obtained a PhD in language testing and assessment, Yuyang Cai is currently working as a Senior Research Assistant in the Sciences of Learning Strategic Research Theme in the Faculty of Education, The University of Hong Kong. His task is to bring together scholars with an interest in the learning sciences from different faculties in order to explore theories, design issues and practice and to improve learning across disciplines. His research interests include the application of a variety of psychometric and statistical techniques (e.g. multidimensional item response theory, generalizability theory, many-faceted Rasch measurement, structural equation modeling, multilevel modeling, growth-curve modeling, and neural network analysis) to the assessment of language proficiency (e.g. learner strategies, reading, grammar, and English for Specific Purposes) and other educational issues.
©2014 by Walter de Gruyter Berlin/Boston
