202 / 2018-08-15 14:59:58

Aligning CSE writing sub-scales with CEFR

CSE; CEFR; Writing; Alignment

全体主题 > 语言测试与语言能力标准的对接研究

This study studies the alignment of CSE writing sub-scales with CEFR. The alignment generally includes content alignment and level alignment. Content alignment aims to seek common ground while reserving differences by addressing the similarities of CEFR and CSE in the general principle and the relation of CEFR categories and CSE categories, which is also the theoretical foundation of CSE-CEFR alignment. Level alignment aims to figure out the relation of CSE nine levels and CEFR six levels based on similar scales and descriptors in CSE and CEFR.
Content alignment is fulfilled by textual comparison and analysis. Level alignment is realized by the large-scale questionnaire survey and interviews. The questionnaires are designed according to the common-item-non-equivalent design (Kolen & Brennan, 2004). All the 467 CEFR descriptors make up of 15 questionnaires, which are linked vertically and horizontally with CEFR internal anchors. And then CSE external anchor descriptors appropriate to the target populations are added to each questionnaire to equate all the CEFR descriptors to the descriptor pool in CSE project. Responses from the teachers and students are respectively analyzed based on Rasch analysis (Linacre, 2014), anchors stability check (Huynh & Meyer, 2010), the fixed common item parameter (FCIP) equating (Taherbhai & Seo, 2013) and a range of statistical analysis.
In terms of content alignment, CSE and CEFR share a lot in common in general principles. Both CSE and CEFR take the communicative approach to language to describe language proficiency from the perspectives of what to do (quantity) and how to do (quality) with the can-do statements. In terms of the level alignment, it is not a one-to-one alignment between the CSE nine levels and the CEFR six broad levels. CSE descriptors of a certain level tend to scatter at several adjacent CEFR levels, one of which is relatively more predominant.